The central challenge for AI agents is not only performance but accountability. Agents that act through opaque prompt sequences may produce correct outputs, but they provide little basis for verifying why an action was permitted, where an error occurred, or how responsibility should be assigned. This paper presents the Structured Cognitive Loop as an architecture for accountable behavior in large language model agents. SCL separates cognition, memory, control, and action into distinct modules. The language model proposes. External memory preserves verified state. A lightweight controller checks preconditions, prevents redundant actions, and authorizes execution before tools are used. We evaluate SCL against ReAct and common LangChain agent variants across travel planning, conditional email drafting, and constraint guided image generation. Across 360 episodes, SCL achieves 86.3 percent task success compared with 70.5 to 76.8 percent for prompt based baselines. It also improves goal fidelity, reduces redundant tool calls, increases reuse of intermediate state, and lowers unsupported assertions. This extended revision situates SCL within a broader architecture of epistemic accountability. Subsequent extensions integrate context aware Human in the Loop control, Pool Gated Retrieval, and the Horizon Warrant Commitment framework. Together these components define an agent architecture in which the model proposes, structure decides, evidence is warranted before use, and human judgment is embedded in the trace rather than imposed after the fact. The result is a foundation for AI agents whose decisions are not only effective but also authorized, inspectable, and accountable.

Real-world user requests to LLM agents are often underspecified. Agents must interact to acquire missing information and make correct downstream decisions. However, current multi-turn GRPO-based methods often rely on trajectory-level reward computation, which leads to credit assignment problems and insufficient advantage signals within rollout groups. A feasible approach is to identify valuable interaction turns at a fine granularity to drive more targeted learning. To address this, we introduce InfoPO (Information-Driven Policy Optimization), which frames multi-turn interaction as a process of active uncertainty reduction and computes an information-gain reward that credits turns whose feedback measurably changes the agent's subsequent action distribution compared to a masked-feedback counterfactual. It then combines this signal with task outcomes via an adaptive variance-gated fusion to identify information importance while maintaining task-oriented goal direction. Across diverse tasks, including intent clarification, collaborative coding, and tool-augmented decision making, InfoPO consistently outperforms prompting and multi-turn RL baselines. It also demonstrates robustness under user simulator shifts and generalizes effectively to environment-interactive tasks. Overall, InfoPO provides a principled and scalable mechanism for optimizing complex agent-user collaboration. Code is available at https://github.com/kfq20/InfoPO.

Agent skills are procedural artifacts that enable LLM agents to execute workflows, verify constraints, and recover from failures. Existing self-evolving methods refine skills using accumulated trajectories. However, they struggle in cold-start settings, where only an initial, imperfect skill is available. Consequently, skill construction defaults to expert authoring or one-shot LLM generation. Expert-authored skills are costly and may not align with how LLM agents actually execute tasks, while one-shot generated skills can be syntactically well formed yet behaviorally weak. To bridge this gap, we propose SkillRevise, an execution-grounded framework designed to iteratively refine these initial skills. SkillRevise diagnoses skill defects from execution evidence, retrieves relevant repair principles from a general memory, and applies execution-anchored edits. By re-executing candidates, it retains the first verifier-passing skill within the revision budget and falls back to empirical utility only when no candidate succeeds. Evaluated across three benchmarks and five LLMs, SkillRevise substantially outperforms one-shot baselines, improving the base agent's success rate on SkillsBench from 36.05% to 61.63%. Furthermore, the revised skills transfer across both executors and task environments, suggesting that SkillRevise captures reusable procedural knowledge beyond any single executor.

AI agent performance is not just a modeling problem, it is fundamentally a systems problem. The advanced capabilities of models are realized through agent harnesses. Therefore, a gap between model assumptions and harness behavior can easily prevent the model's full capabilities from translating into agent performance. We formalize this as the `intent-execution' gap: the mismatch between what the model intends and what the harness executes, and vice versa. We argue that minimizing this intent-execution gap is as important as other aspects of harness design such as tools and execution loops. To illustrate the impact of this harness-model alignment, we develop a simple and customizable harness called `Simple Strands Agent' (SSA). SSA aims to find the bulk of common patterns which generalize across different model families (such as Claude, Gemini, GPT, Grok, Qwen), as well as a small number of model-specific preferences. We make two contributions: (i) we reproduce or improve on the pass@1 performance reported by diverse model-provider families on popular agentic benchmarks (SWE-Pro, SWE-Verified and Terminal-Bench-2), and (ii) building on an analysis of 138k trajectories generated by SSA, we look beyond the pass@1 numbers which tend to be relatively even across frontier models. By representing agent trajectories in code state-spaces, we observe model-level differences in problem-solving behavior. Finer-grained metrics such as edit frequency, testing activity, and phase-transitions reveal how individual models allocate effort across different stages of autonomous problem solving.

Memory management is essential for LLM agents in long-term interactions. Current memory frameworks typically treat agents as passive ``recorders'' and retrieve information without understanding its deeper implications. They may fail in scenarios requiring reasoning and complex decision-making. To bridge this critical gap, we propose a novel actionable memory framework called ActMem that integrates memory retrieval with active causal reasoning. ActMem transforms unstructured dialogue history into a structured causal and semantic graph. By leveraging counterfactual reasoning and commonsense completion, it enables agents to deduce implicit constraints and resolve potential conflicts between past states and current intentions. Furthermore, we introduce a comprehensive dataset ActMemEval to evaluate agent reasoning capabilities in logic-driven scenarios, moving beyond the fact-retrieval focus of existing memory benchmarks. Experiments demonstrate that ActMem significantly outperforms baselines in handling complex, memory-dependent tasks, paving the way for more consistent and reliable intelligent assistants.

LLM-based shopping agents increasingly rely on long purchase histories and multi-turn interactions for personalization, yet naively appending raw history to prompts is often ineffective due to noise, length, and relevance mismatch. We propose MemRerank, a preference memory framework that distills user purchase history into concise, query-independent signals for personalized product reranking. To study this problem, we build an end-to-end benchmark and evaluation framework centered on an LLM-based \textbf{1-in-5} selection task, which measures both memory quality and downstream reranking utility. We further train the memory extractor with reinforcement learning (RL), using downstream reranking performance as supervision. Experiments with two LLM-based rerankers show that MemRerank consistently outperforms no-memory, raw-history, and off-the-shelf memory baselines, yielding up to \textbf{+10.61} absolute points in 1-in-5 accuracy. These results suggest that explicit preference memory is a practical and effective building block for personalization in agentic e-commerce systems.

Text-agent environments are typically modeled as partially observable Markov decision processes (POMDPs), assuming that the simulator's latent state and transition dynamics are hidden from the agent. Yet little work has examined whether executable code can be induced to serve as a world model for prediction and planning under partial observability. We introduce PatchWorld, a gradient-free framework that turns offline trajectories into executable Python world models through counterexample-guided code repair. Instead of predicting the next observation with a black-box model, PatchWorld induces symbolic belief-state programs whose action updates can be inspected, replayed, and locally patched. Across seven AgentGym environments, PatchWorld-Simple achieves the highest code-based planning score among evaluated methods, reaching 76.4\% macro success in live one-step lookahead while invoking no LLM calls inside the world-model prediction module itself. We further find that a human-specified residual-memory bias improves surface observation fidelity but weakens decision utility. This exposes a tradeoff in executable world models, since improving observation fidelity can come at the expense of action-discriminative dynamics, and vice versa. Code is available at https://github.com/HKBU-KnowComp/PatchWorld.

Multi-hop logical reasoning on knowledge graphs requires faithfully mapping the logical semantics to latent space. Current geometric embedding methods show to be useful on this task by mapping entities to geometric regions and logical operations to latent transformations. While a geometric embedding can provide a direct interpretability framework for query answering, current methods have only leveraged the geometric construction of entities, failing to map logical operations to pure geometric transformations and, instead, using neural components to learn these operations. On the other hand, purely neural-based methods outperform geometric methods, but they lack interpretability in the latent space. We introduce GeometrE, a geometric embedding method for multi-hop reasoning, that maps every logical operation to a purely geometric operation in the latent space. Additionally, we introduce a transitive loss function and show that, unlike existing methods, it can preserve the logical rule for all a,b,c: r(a,b) and r(b,c) -> r(a,c). Our experiments show that GeometrE outperforms current state-of-the-art geometric methods and remains competitive with existing neural-based methods on standard benchmark datasets.

Geometric problem solving, as a typical multimodal reasoning problem, has attracted much attention and made great progress recently, however most of works focus on plane geometry while usually fail in solid geometry due to 3D spatial diagrams and complex reasoning. To bridge this gap, we introduce Hilbert-Geo, the first unified formal language framework for solid geometry, including an extensive predicate library and a dedicated theorem bank. Based on this framework, we propose a Parse2Reason method containing two steps of first parsing then reasoning. In the parsing step, we utilize conditional description language (CDL), a formalized language composed of predicates specifically designed to construct geometric conditions, to represent both problem description (natural text) and solid diagrams (visual image). In the reasoning step, we leverage those formal CDL and the theorem bank to perform relational inference and algebraic computation, generating strictly correct, verifiable, and human-readable reasoning processes. Notably, our proposed Hilbert-Geo is also applicable to plane geometry. To advance geometric reasoning, we curate two expert-annotated dataset SolidFGeo2k and PlaneFGeo3k, which are furnished with geometric formal language annotations, solutions and answers. Extensive experiments show that our proposed method achieves the state-of-the-art (SOTA) performance 77.3% in SolidFGeo2k and 84.1% in MathVerse-Solid (one small subset in MathVerse dedicated to solid geometry), substantially outperforming leading MLLMs, such as Gemini-2.5-pro (54.2% on SolidFGeo2k) and GPT-5 (62.9% on MathVerse-Solid). In addition, our method achieves the SOTA accuracy 80.2% in PlaneFGeo3k, demonstrating the generality of the Hilbert-Geo in geometric reasoning. Our code and datasets are released at https://github.com/PremiLab-Math/Hilbert-Geo.

Reinforcement learning under partial observability requires deciding what information to retain, yet most memory-based approaches do not explicitly model short-term-to-long-term transfer of symbolic observations. We study this transfer process in a temporal knowledge-graph memory setting and cast it as a neuro-symbolic value-based decision problem: for each observed triple, the agent chooses whether to keep or drop it before long-term insertion. To handle variable-sized short-term buffers, we use a per-item Q-learning design with shared parameters and a practical temporal-difference update over matched items across consecutive steps. On the RoomKG benchmark at long-term memory capacity 128, learned transfer decisions outperform symbolic and neural baselines, including symbolic baselines with temporal annotations and history-based LSTM/Transformer baselines. Across transfer-policy ablations, a lightweight local short-term-only variant performs best, and step-level behavior shows that the policy keeps navigation- and query-relevant facts while discarding lower-value candidate facts, supporting explicit and interpretable memory decisions under memory constraints.

TLA+ is a formal specification language for verifying distributed systems and safety-critical protocols. Large language models (LLMs) frequently produce TLA+ specifications that fail the TLC model checker for semantic reasons. Across 25 LLMs, the best public baseline is 26.6% syntactic parse and 8.6% semantic model-check. We present TLA-Prover, a 20-billion-parameter model for TLA+ specification synthesis. Training combines supervised fine-tuning (SFT) on verified examples with repair-based group-relative policy optimization (GRPO). In the GRPO stage, the model learns to fix its own rejected specifications. We also train a direct preference optimization (DPO) variant from the same SFT checkpoint as an ablation. TLC provides the reward signal directly, with no learned reward model. Four tiers grade each output: Bronze (parses), Silver (no warnings), Gold (passes TLC), and Diamond. To reach Diamond, the model's correctness property is automatically altered in a small way; TLC must then detect a violation. If TLC still passes, the property was always-true and contributes nothing; the output fails Diamond. TLA-Prover reaches 9/30 (i.e. pass@1 = 30%) at both Gold and Diamond on a held-out 30-problem benchmark. This is roughly 3.5x the 8.6% untuned baseline. The DPO variant reaches 20% at Diamond. Gold and Diamond coincide at every checkpoint; this prevents the trivial-property failure mode.

Game-theoretic dynamics between AI agents could differ from traditional human-human interactions in various ways. One such difference is that it may be possible to accurately simulate an AI agent, for example because its source code is known. Such an agent would then be fundamentally uncertain whether it is in the real world or in a simulation. Our aim is to explore ways of leveraging this possibility to achieve more cooperative outcomes in strategic settings. In this paper, we study an interaction between AI agents where the agents run a recursive joint simulation. That is, the agents first jointly observe a simulation of the situation they face. This simulation in turn recursively includes additional simulations (with a small chance of failure, to avoid infinite recursion), and the results of all these nested simulations are observed before an action is chosen. We show that the resulting interaction is strategically equivalent to an infinitely repeated version of the original game, allowing a direct transfer of existing results such as the various folk theorems. As evidence that the equivalence is robust, we show that it holds even when we relax some of the assumptions and that it also holds ``from the inside'' -- meaning, for an agent that finds itself inside the game and has self-locating uncertainty.

Automating scientific poster generation requires hierarchical document understanding and coherent content-layout planning. Existing methods often rely on flat summarization or optimize content and layout separately. As a result, they often suffer from information loss, weak logical flow, and poor visual balance. We present PosterForest, a training-free framework for scientific poster generation. Our method introduces the Poster Tree, a structured intermediate representation that captures document hierarchy and visual-textual semantics across multiple levels. Building on this representation, content and layout agents perform hierarchical reasoning and recursive refinement, progressively optimizing the poster from global organization to local composition. This joint optimization improves semantic coherence, logical flow, and visual harmony. Experiments show that PosterForest outperforms prior methods in both automatic and human evaluations, without additional training or domain-specific supervision.

In recent years, the advances of large language models and autonomous agents have revolutionized the healthcare field, facilitating diagnosis and improving treatment results. However, most existing AI systems rely on pre-trained knowledge and predefined pipelines, which struggle to learn dynamically from the interactive chat session history that contains patient outcomes and past failures. To address this limitation, we propose VIBEMed, a multi-agent framework with a built-in self-evolution mechanism and architecture-level safety sandbox for robust clinical decision support. The system integrates three specialized agents, including a Clinical Diagnostic Agent (CDA) for hypothesis generation, a Therapeutic Execution Agent (TEA) for treatment planning, and a Clinical Evolution Manager Agent (CEMA) that distills longitudinal clinical feedback into reusable knowledge, transforming multimodal patient information into personalized medical decisions. Through self-evolution mechanism, the framework enables iterative updates across memory, model behavior, and decision strategies, allowing the system to improve over time. Experimental results show that VIBEMed demonstrates superior performance through its evolving mechanism in complex clinical cases, particularly in tasks that require integrated decision-making and longitudinal planning. The framework also supports reliable end-to-end decisions in challenging scenarios such as oncology treatment planning, highlighting its feasibility in real-world clinical contexts. Overall, VIBEMed provides a practical path beyond static AI systems toward adaptive, experience-driven clinical decision support, demonstrating the value of combining multi-agent collaboration with continuous evolution for advancing precision medicine.

Leveraging multiple Large Language Models (LLMs) has proven effective for addressing complex, high-dimensional tasks, but current approaches often rely on static, manually engineered multi-agent configurations. To overcome these constraints, we present the Agentic Neural Network (ANN), a framework that conceptualizes multi-agent collaboration as a layered neural network architecture. In this design, each agent operates as a node, and each layer forms a cooperative team focused on a specific subtask. Our framework follows a two-phase optimization strategy: (1) Forward Phase - Drawing inspiration from neural network forward passes, tasks are dynamically decomposed into subtasks, and cooperative agent teams with suitable aggregation methods are constructed layer by layer. (2) Backward Phase - Mirroring backpropagation, we refine both global and local collaboration through iterative feedback, allowing agents to self-evolve their roles, prompts, and coordination. This neuro-symbolic approach enables our framework to create new or specialized agent teams post-training, delivering notable gains in accuracy and adaptability. Across seven benchmark datasets, our work surpasses leading multi-agent baselines under the same configurations, showing consistent performance improvements.

Imitation Learning (IL) is a natural way for humans to teach robots, particularly when high-quality demonstrations are easy to obtain. While IL has been widely applied to single-robot settings, relatively few studies have addressed the extension of these methods to multi-agent systems, especially in settings where a single human must provide demonstrations to a team of collaborating robots. In this paper, we introduce and study Round-Robin Behavior Cloning (R2BC), a method that enables a single human operator to effectively train multi-robot systems through sequential, single-agent demonstrations. Our approach allows the human to teleoperate one agent at a time and incrementally teach multi-agent behavior to the entire system, without requiring demonstrations in the joint multi-agent action space. We show that R2BC methods match, and in some cases surpass, the performance of an oracle behavior cloning approach trained on privileged synchronized demonstrations across four multi-agent simulated tasks. Finally, we deploy R2BC on two physical robot tasks trained using real human demonstrations.

Recent studies have suggested that Large Language Models (LLMs) could provide interesting ideas contributing to mathematical discovery. This claim was motivated by reports that LLM-based genetic algorithms produced heuristics offering new insights into the online bin packing problem under uniform and Weibull distributions. In this work, we reassess this claim through a detailed analysis of the heuristics produced by LLMs, examining both their behavior and interpretability. Despite being human-readable, these heuristics remain largely opaque even to domain experts. Building on this analysis, we propose a new class of algorithms tailored to these specific bin packing instances. The derived algorithms are significantly simpler, more efficient, more interpretable, and more generalizable, suggesting that the considered instances are themselves relatively simple. We then discuss the limitations of the claim regarding LLMs' contribution to this problem, which appears to rest on the mistaken assumption that the instances had previously been studied. Our findings instead emphasize the need for rigorous validation and contextualization when assessing the scientific value of LLM-generated outputs.

The Capacitated Vehicle Routing Problem (CVRP), a fundamental combinatorial optimization challenge, focuses on optimizing fleet operations under vehicle capacity constraints. While extensively studied in operational research, the NP-hard nature of CVRP continues to pose significant computational challenges, particularly for large-scale instances. This study presents AILS-AHD (Adaptive Iterated Local Search with Automatic Heuristic Design), a novel approach that leverages Large Language Models (LLMs) to revolutionize CVRP solving. Our methodology integrates an evolutionary search framework with LLMs to dynamically generate and optimize ruin heuristics within the AILS method. Additionally, we introduce an LLM-based acceleration mechanism to enhance computational efficiency. Comprehensive experimental evaluations against state-of-the-art solvers, including AILS-II and HGS, demonstrate the superior performance of AILS-AHD across both moderate and large-scale instances. Notably, our approach establishes new best-known solutions for 8 out of 10 instances in the CVRPLib large-scale benchmark, underscoring the potential of LLM-driven heuristic design in advancing the field of vehicle routing optimization.

Heuristic search is the dominant paradigm in symbolic AI planning, and the strongest heuristics are the result of decades of work by planning researchers. Recent work has shown that large language models (LLMs) can design heuristics for individual planning domains, but no LLM-generated heuristic has so far worked on arbitrary planning tasks. In this paper, we use evolutionary search to produce the first LLM-generated domain-independent heuristics that exceed the hand-engineered state of the art. We let an LLM mutate parent heuristics written in C++, store candidates in a MAP-Elites archive keyed on informedness and speed and calculate fitness scores by blending coverage with solving time. To place the evolved programs in context, we additionally benchmark a broad set of hand-engineered heuristics on their informedness-speed tradeoff, which to our knowledge has not been done before. On unseen testing domains, our best evolved heuristic solves more tasks than even the strongest baseline, with our full heuristic suite spanning the Pareto frontier of said tradeoff. We also find that seeding evolution from the trivial blind heuristic outperforms seeding from the strong FF heuristic, even when the resulting program is itself an FF variant, and that LLM reasoning effort affects how often candidates compile much more than the quality of those that do. Because the evolved programs are plain C++, they slot into existing planners as drop-in replacements and inherit the soundness and completeness guarantees of the underlying search.

Extending Bayesian optimization to batch evaluation can enable the designer to make the most use of parallel computing technology. However, most of current batch approaches do not scale well with the batch size. That is, their optimization efficiencies often deteriorate as the batch size increases. To address this issue, we propose a simple and efficient approach to extend Bayesian optimization to large-scale batch evaluation in this work. Different from existing batch approaches, the idea of the new approach is to draw a batch of axis-aligned subspaces of the original problem and select one point from each subspace using existing acquisition functions. Numerical experiments show that our proposed approach speedups the convergence significantly when compared with the sequential Bayesian optimization algorithm, and performs very competitively when compared with ten batch Bayesian optimization algorithms. The implementation of our proposed approach is available at https://github.com/zhandawei/SubSpace_Acquisition_Functions.

Large Language Models (LLMs) have advanced Automatic Heuristic Design (AHD) by enabling heuristic generation through reasoning and code synthesis. Existing LLM-based AHD architectures mainly follow two paradigms: Natural Evolution, which uses crossover and mutation to explore heuristic programs, and Metacognitive Evolution, which refines reasoning through reflection. However, Natural Evolution discards reasoning traces, weakening knowledge inheritance and exploitation, while Metacognitive Evolution lacks population-level recombination, limiting exploration and increasing the risk of premature convergence. These limitations reduce search efficiency, stability, and solution quality on complex problems. To address this gap, we propose MeEvo, a dual-layer AHD framework that cyclically couples Natural Evolution and Metacognitive Evolution. Natural Evolution explores heuristic code while recording reasoning traces, fitness values, and errors into a shared history; Metacognitive Evolution then reflects on this history to generate improved heuristics that re-enter the parent pool for the next cycle. This design enables population-driven exploration and reflection-driven refinement to reinforce each other. Experiments on five optimization problems with two LLM backbones show that MeEvo achieves stronger and more stable performance than existing LLM-based AHD architectures, especially on complex constrained tasks.

State Space Models (SSMs) have emerged as an efficient alternative to the Transformer architecture. Prior work shows that, when trained under comparable conditions, SSMs can match or surpass Transformers on code understanding tasks. However, their internal mechanisms remain a black box. We present the first systematic analysis of what SSM-based code models learn along with the direct comparison between SSM and Transformer models in this domain. Our analysis shows that SSMs capture syntactic and semantic structure more effectively than Transformers during pretraining but forgets certain relations during fine-tuning on some tasks. To investigate this behavior, we introduce SSM-Interpret, a frequency-domain framework that exposes a spectral shift toward short-range dependencies during fine-tuning. Guided by these findings, we propose architectural modifications that significantly improve the performance of SSM-based code model by upto +6 MRR on NLCodeSearch. This demonstrates that our analysis not only explains model behavior but also leads directly to better designs.

Offline reinforcement learning (RL) enables data-efficient and safe policy learning without online exploration, but its performance often degrades under distribution shift. The learned policy may visit out-of-distribution state-action pairs where value estimates and learned dynamics are unreliable. To address policy-induced extrapolation and transition uncertainty in a unified framework, we formulate offline RL as robust policy optimization, treating the transition kernel as a decision variable within an uncertainty set and optimizing the policy against the worst-case dynamics. We propose Robust Regularized Policy Iteration (RRPI), which replaces the intractable max-min bilevel objective with a tractable KL-regularized surrogate and derives an efficient policy iteration procedure based on a robust regularized Bellman operator. We provide theoretical guarantees by showing that the proposed operator is a $γ$-contraction and that iteratively updating the surrogate yields monotonic improvement of the original robust objective with convergence. Experiments on D4RL benchmarks demonstrate that RRPI achieves strong average performance, outperforming recent baselines including percentile-based methods on the majority of environments while remaining competitive on the rest. Moreover, RRPI exhibits robust performance by aligning lower $Q$-values with high epistemic uncertainty, which prevents the policy from executing unreliable out-of-distribution actions.

Current Large Reasoning Models (LRMs) exhibit remarkable general capabilities but significantly underperform in spatial reasoning tasks. Existing approaches treat this gap as a knowledge deficit, relying on supervised fine-tuning (SFT) to ingest labeled spatial data from external vision sources or synthetic engines. In contrast, we argue that for many tasks, spatial reasoning capabilities are already present in pre-trained LRMs but require alignment through logical coherence under geometric 2D and 3D constraints. In this work, we propose a self-supervised reinforcement learning (RL) framework that targets the internal reasoning process without requiring ground-truth annotations. By formalizing the notion of consistency verifiers -- reward functions that check for geometric and semantic consistency under transformations -- we demonstrate that models can improve their spatial reasoning abilities. We use both image transformations, like flipping, and textual transformations, like swapping the order of objects in the question, and propose a new optimal transport-based RL strategy, OT-GRPO, which is a minimal-matching variant of group relative policy optimization tailored to pairwise verifiers. We show that this label-free consistency training approaches the accuracy of models trained with ground-truth supervision and achieves similar generalization across diverse tasks and data domains.

Graphical user interface (GUI) grounding requires vision-language models (VLMs) to identify small target elements in high-resolution screenshots and predict precise screen coordinates. On-policy self-distillation (OPSD) is a promising post-training approach for this coordinate-sensitive task, since it provides dense token-level teacher signals beyond hard coordinate labels. However, naive OPSD is not well suited to GUI grounding: OPSD evaluates the teacher on student-generated prefixes, the quality of coordinate-token teacher signals can degrade when the prefix has already deviated from the target coordinate, leading to unreliable teacher signal. To mitigate this, We propose quality-aware self-distillation for VLM-based GUI grounding, which improves coordinate-token teacher-signal quality through soft correctness-aware gating and teacher-probability scaling. The soft correctness-aware gate checks whether the teacher's current coordinate-token prediction can still be completed into the ground-truth box under the student-generated prefix. If not, the corresponding teacher signal is down-weighted. Teacher-probability scaling then uses the teacher's confidence as a lightweight factor to further calibrate the strength of the gated supervision. A key empirical finding is that neither component alone improves overall performance, whereas combining them consistently improves performance. This suggests that the two mechanisms play complementary roles: correctness-aware gating suppresses unreliable coordinate-token supervision, while teacher-probability scaling calibrates the strength of the remaining signals. Experiments across six GUI grounding benchmarks show that our method consistently improves the base model and outperforms strong baselines.

Federated Learning (FL) is a promising paradigm for finetuning Large Language Models (LLMs) across distributed data sources while preserving data privacy. However, finetuning such large models is challenging on edge devices due to its high resource demand. Zeroth-order Optimization (ZO) estimates gradients through finite-difference approximations, which rely on function evaluations under random perturbations of the model parameters. Consequently, ZO with task alignment provides a potential solution, allowing finetuning using only forward passes with inference-level memory requirements and low communication overhead, but it suffers from slow convergence and higher computational demand. In this paper, we propose a new ZO-based method that applies a more efficient technique to reduce the computational demand associated with using a large number of perturbations while preserving their convergence benefits. This is achieved by splitting the model into consecutive blocks and allocating a higher number of perturbations to the second block, enabling efficient reuse of intermediate activations to update the full network with fewer forward evaluations. Our evaluation on RoBERTa-large, OPT1.3B, LLaMa-3-3.2B models shows up to $3\times$ reduction in computation compared to the other ZO-based techniques, while retaining the memory and communication benefits over first-order federated learning techniques.

Transformers have revolutionized the field of machine learning. In particular, they can be used to solve complex algorithmic problems, including graph-based tasks. In such algorithmic tasks a key question is what is the minimal size of a transformer that can implement the task. Recent work has begun to explore this problem for graph-based tasks, showing that for sub-linear embedding dimension (i.e., model width) logarithmic depth suffices. However, an open question, which we address here, is what happens if width is allowed to grow linearly, while depth is kept fixed. Here we analyze this setting, and provide the surprising result that with linear width, constant depth suffices for solving a host of graph-based problems. This suggests that a moderate increase in width can allow much shallower models, which are advantageous in terms of inference and train time. For other problems, we show that quadratic width is required. Our results demonstrate the complex and intriguing landscape of transformer implementations of graph-based algorithms. We empirically investigate these trade-offs between the relative powers of depth and width and find tasks where wider models have the same accuracy as deep models, while having much faster train and inference time due to parallelizable hardware.

In this paper, we delve deeper into the Kullback-Leibler (KL) Divergence loss and mathematically prove that it is equivalent to the Decoupled Kullback-Leibler (DKL) Divergence loss that consists of (1) a weighted Mean Square Error (wMSE) loss and (2) a Cross-Entropy loss incorporating soft labels. Thanks to the decoupled structure of DKL loss, we have identified two areas for improvement. Firstly, we address the limitation of KL loss in scenarios like knowledge distillation by breaking its asymmetric optimization property along with a smoother weight function. This modification effectively alleviates convergence challenges in optimization, particularly for classes with high predicted scores in soft labels. Secondly, we introduce class-wise global information into KL/DKL to reduce bias arising from individual samples. With these two enhancements, we derive the Generalized Kullback-Leibler (GKL) Divergence loss and evaluate its effectiveness by conducting experiments on CIFAR-10/100, ImageNet, and vision-language datasets, focusing on adversarial training, and knowledge distillation tasks. Specifically, we achieve new state-of-the-art adversarial robustness on the public leaderboard -- RobustBench and competitive knowledge distillation performance across CIFAR/ImageNet models and CLIP models, demonstrating the substantial practical merits. Our code is available at https://github.com/jiequancui/DKL.

Implicit Neural Representations (INRs) have recently shown impressive results, but their fundamental capacity, implicit biases, and scaling behavior remain poorly understood. We investigate the performance of diverse INRs across a suite of 2D and 3D real and synthetic signals with varying effective bandwidth, as well as both overfitting and generalization tasks including tomography, super-resolution, and denoising. By stratifying performance according to model size as well as signal type and bandwidth, our results shed light on how different INR and grid representations allocate their capacity. We find that, for many tasks involving dense signals, a simple regularized grid with interpolation trains faster and to higher or comparable quality than any INR with the same number of parameters. We also find limited settings -- namely fitting binary signals such as shape contours -- where INRs outperform grids, to guide future development and use of INRs towards the most advantageous applications.

Modern deep learning is increasingly characterized by the use of open-weight foundation models that can be fine-tuned on specialized datasets. This has led to a proliferation of expert models and adapters, often shared via platforms like HuggingFace and AdapterHub. Model merging has recently emerged as an effective way to leverage these existing resources, enabling the composition of capabilities from different model checkpoints. A natural pipeline has thus formed to harness the benefits of transfer learning and amortize sunk training costs: models are pre-trained on general data, fine-tuned on specific tasks, and then multiple checkpoints are merged to obtain a more capable model. A prevailing assumption is that improvements at one stage of this pipeline propagate downstream, leading to gains at subsequent steps. In this work, we challenge that assumption by examining how expert fine-tuning affects model merging. We show that long fine-tuning of experts that optimizes for their individual performance leads to degraded merging performance across vision and language modalities, multiple model scales, and both fully fine-tuned and LoRA-adapted models. We trace this degradation to the memorization of a small set of difficult examples that dominate late fine-tuning steps. This causes negative parameter interference and encodes knowledge that is forgotten during merging. Finally, we demonstrate that task-dependent aggressive early stopping strategies can significantly improve model merging performance.

Post Training Quantization (PTQ), a mainstream model compression technique, often leads to the paradoxical 'low error, high loss' phenomenon because it focuses solely on minimizing quantization error. The root cause lies in the Hessian matrix of the LLM loss landscape: a few high curvature directions are extremely sensitive to perturbations. To address this, we propose the Hessian Robust Quantization (HeRo Q) algorithm, which applies a lightweight, learnable rotation-compression matrix to the weight space prior to quantization. This joint framework reshapes the loss landscape by reducing the largest Hessian eigenvalue and reducing its max eigenvalue, thereby significantly enhancing robustness to quantization noise. HeRo-Q requires no architectural modifications, incurs negligible computational overhead, and integrates seamlessly into existing PTQ pipelines. Experiments on Llama and Qwen models show that HeRo Q consistently outperforms state of the art methods including GPTQ, AWQ, and SpinQuant not only achieving superior performance under standard W4A8 settings, but also excelling in the highly challenging W3A16 ultra low bit regime, where it boosts GSM8K accuracy on Llama3 8B to 70.15\% and effectively avoids the logical collapse commonly seen in aggressive quantization.

In this paper, we formulate the compression of large language models (LLMs) by optimally deleting transformer blocks (``block removal'') as a constrained binary optimization (CBO) problem that can be mapped to a physical system (Ising glass), whose energies are a strong proxy for downstream model performance. This formulation enables an efficient ranking of a large number of candidate block-removal configurations yielding many high-quality, non-trivial solutions beyond those only removing consecutive regions. Our method performs strongly in the deep compression regime, such as for 50% compression of Llama-3.3-70B-Instruct, where we achieve an almost 23 percentage point increase on the MMLU benchmark compared to other state-of-the-art (SOTA) block-removal methods. For lighter compression, it performs on par with those methods across several benchmarks for Llama-3.1-8B-Instruct, Qwen3-14B (both before and after retraining), as well as Llama-3.3-70B-Instruct. The approach is computationally efficient and requires only forward and backward passes on a calibration dataset for a few active parameters. Additionally, we demonstrate that using good heuristic solvers for the CBO problem provides solutions that perform well on downstream tasks in negligible runtime when it is unfeasible to solve the problem exactly. The method can be readily applied to any architecture. We illustrate this generality on the recent NVIDIA-Nemotron-3-Nano-30B-A3B-FP8 model, which exhibits a highly inhomogeneous and challenging block structure, and where we outperform SOTA for AIME25 and GPQA when removing either 2 attention layers or 3 mixture-of-experts layers.

Diffusion posterior sampling solves inverse problems by combining a pretrained diffusion prior with measurement-consistency guidance. However, full-band guidance can be unreliable at high noise levels, where clean estimates contain score-induced errors and high-frequency measurement directions are weakly identifiable. We argue that posterior guidance should expose measurement frequencies according to the instantaneous diffusion noise level. Based on this principle, we propose a posterior continuation framework that constructs a family of intermediate posteriors whose likelihood emphasizes currently reliable frequency bands and gradually returns to full-band consistency. We instantiate this framework with a stabilized sampler that combines a diffusion predictor, frequency-limited likelihood refinement, and a Haar-domain commitment rule that commits reliable coarse corrections while deferring weakly identifiable details. Across super-resolution, inpainting, and deblurring, our method achieves competitive-to-state-of-the-art restoration performance, including up to 5 dB PSNR improvement on motion deblurring over strong baselines in evaluations on FFHQ and ImageNet.

Continual learning has become a trending topic in machine learning. Recent studies have discovered an interesting phenomenon called loss of plasticity, referring to neural networks gradually losing the ability to learn new tasks. However, existing plasticity research largely relies on benchmarks with abrupt task transitions, without examining whether the abruptness itself contributes to the observed plasticity loss. In this paper, we investigate the role of transition abruptness by simulating gradually changing environments through input/output interpolation and task sampling. We perform theoretical and empirical analysis, showing that the severity of plasticity loss is closely tied to the abruptness of task transitions, and can be substantially reduced when the environment changes gradually.

Dysarthric speech quality assessment (DSQA) is critical for clinical diagnostics and inclusive speech technologies. However, subjective evaluation is costly and difficult to scale, and the scarcity of labeled data limits robust objective modeling. To address this, we propose a three-stage framework that leverages unlabeled dysarthric speech and large-scale typical speech datasets to scale training. A teacher model first generates pseudo-labels for unlabeled samples, followed by weakly supervised pretraining using a label-aware contrastive learning strategy that exposes the model to diverse speakers and acoustic conditions. The pretrained model is then fine-tuned for the downstream DSQA task. Experiments on five unseen datasets spanning multiple etiologies and languages demonstrate the robustness of our approach. Our Whisper-based baseline significantly outperforms SOTA DSQA predictors such as SpICE, and the full framework achieves an average SRCC of 0.761 across unseen test datasets.

Grokking in transformers trained on algorithmic tasks is characterized by a long delay between training-set fit and abrupt generalization, but the source of that delay remains poorly understood. In encoder-decoder arithmetic models, we argue that this delay reflects limited access to already learned structure rather than failure to acquire that structure in the first place. We study one-step Collatz prediction and find that the encoder organizes parity and residue structure within the first few thousand training steps, while output accuracy remains near chance for tens of thousands more. Causal interventions support the decoder bottleneck hypothesis. Transplanting a trained encoder into a fresh model accelerates grokking by 2.75 times, while transplanting a trained decoder actively hurts. Freezing a converged encoder and retraining only the decoder eliminates the plateau entirely and yields 97.6% accuracy, compared to 86.1% for joint training. What makes the decoder's job harder or easier depends on numeral representation. Across 15 bases, those whose factorization aligns with the Collatz map's arithmetic (e.g., base 24) reach 99.8% accuracy, while binary fails completely because its representations collapse and never recover. The choice of base acts as an inductive bias that controls how much local digit structure the decoder can exploit, producing large differences in learnability from the same underlying task.

We present Clin-JEPA, a multi-phase co-training framework for joint-embedding predictive (JEPA) pretraining on EHR patient trajectories. JEPA architectures have enabled latent-space planning in robotics and high-quality representation learning in vision, but extending the paradigm to EHR data -- to obtain a single backbone that simultaneously forecasts patient trajectories and serves diverse downstream risk-prediction tasks without per-task fine-tuning -- remains an open challenge. Existing JEPA frameworks either discard the predictor after pretraining (I-JEPA, V-JEPA) or train it on a frozen pretrained encoder (V-JEPA 2-AC), leaving the encoder unaware of the rollout signal that the retained predictor must use at inference; co-training the encoder and predictor under a shared JEPA prediction objective would supply this grounding, but naïve co-training is unstable, with representation collapse and online/target drift causing autoregressive rollout to diverge. Clin-JEPA's five-phase pretraining curriculum -- predictor warmup, joint refinement, EMA target alignment, hard sync, and predictor finalization -- addresses each failure mode by phase, stably co-training a Qwen3-8B-based encoder and a 92M-parameter latent trajectory predictor. On MIMIC-IV ICU data, three independent evaluations support the framework: (1) latent $\ell_1$ rollout drift uniquely converges ($-$15.7%) over 48-hour horizons while baselines and ablations diverge (+3% to +4951%); (2) the encoder learns a clinically discriminative latent geometry (deteriorating-patient cohorts displace 4.83$\times$ further than stable patients in latent space, vs $\leq$2.62$\times$ for baseline encoders); (3) a single backbone outperforms strong tabular and sequence baselines on multi-task downstream evaluation. Clin-JEPA achieves mean AUROC 0.851 on ICareFM EEP and 0.883 on 8 binary risk tasks (+0.038 and +0.041 vs baseline average).

A persistent paradox in time-series forecasting is that structurally simple MLP and linear models often outperform high-capacity Transformers. We argue that this gap arises from a mismatch in the sequence-modeling primitive: while many time-series dynamics are governed by global temporal operators (e.g., filtering and harmonic structure), standard attention forms each output as a convex combination of inputs. This restricts its ability to represent signed and oscillatory transformations that are fundamental to temporal signal processing. We formalize this limitation as a simplex-constrained mixing bottleneck in softmax attention, which becomes especially restrictive for operator-driven time-series tasks. To address this, we propose $\textbf{Temporal Operator Attention (TOA)}$, a framework that augments attention with explicit, learnable sequence-space operators, enabling direct signed mixing across time while preserving input-dependent adaptivity. To make dense $N \times N$ operators practical, we introduce Stochastic Operator Regularization, a high-variance dropout mechanism that stabilizes training and prevents trivial memorization. Across forecasting, anomaly detection, and classification benchmarks, TOA consistently improves performance when integrated into standard backbones such as PatchTST and iTransformer, with particularly strong gains in reconstruction-heavy tasks. These results suggest that explicit operator learning is a key ingredient for effective time-series modeling.

In the field of quantum reservoir computing (QRC), many different computational models and architectures have been proposed. From these models, we identify feedback-based models -- which use a feedback mechanism to re-embed classical measurements from the QRC -- and recurrent models -- which use a multi-register approach with memory and readout qubits -- as the two major competing architectures that have been discussed and validated on hardware. In this paper, we advance upon the recurrent architectures, which employ a two register approach to endow the QRC with a fading memory. While these approaches have been validated on hardware and have demonstrated great real-world performance on noisy-intermediate-scale-quantum (NISQ) quantum processing units (QPUs), the exact mechanism through which the memory capacity arises is not completely understood or fully controllable. With this, we augment the recurrent approaches and present a hardware-realizable mechanism, which we call a tunable partial-SWAP, that allows for the direct control of the rate of memory dissipation from a QRN implemented on a gate-based QPU. The theory behind this mechanism is discussed in terms of a controlled amplitude-damping channel and validation experiments using a randomized short-term memory capacity (STMC) recall benchmark and the NARMA-5 dataset are conducted using simulation and IBM QPUs, respectively.

Block-level residual routing makes learned residual aggregation practical by routing over block summaries, but each summary compresses an ordered sequence of attention and MLP updates into one cumulative vector. We propose \method{}, a lightweight residual basis router that keeps the cumulative block source and adds one half-split detail basis, computed as the difference between first-half and second-half residual updates. The detail basis is RMS-matched and updated online, exposing coarse intra-block trajectory information without dense sublayer-level routing. Across OpenWebText, cross-domain character-level benchmarks, and BPE-tokenized OpenWebText, the empirical pattern is depth-dependent: gains are small or mixed at shallow depth and most reliable in 48-layer models. In the 201M 48-layer setting, \method{} improves over Block AttnRes across all three seeds, while a 453M two-seed probe shows the same direction. Ablations rule out source duplication, random signed details, fixed detail-source biases, or block-count changes alone. Cost analysis shows that the method is FLOP-light but not wall-clock-free: it adds memory and routing overhead, yet its relative arithmetic cost is amortized as width grows and earlier convergence can reduce time-to-target.

In time-series generation, existing approaches typically handcraft ortrain a separate model for each dataset, which hinders their scalability and fails to leverage shared temporal structures across domains. To address this fragmentation, we propose UPLOTS, a Unified, Prompt-guided Language model framework fOr constrained Time-Series Generation across diverse domains. Instead of building task-specific models, UPLOTS leverages a single pre-trained transformer backbone guided by learned constraint prompts, enabling on-demand generation with precise pattern control. One key innovation is our dynamic multi-dataset loss re-weighting and prompt-to-pattern mapping, which allows UPLOTS to internalize diverse temporal structures during training and conditionally generate them at inference. We evaluate UPLOTS on four real-world benchmarks and multiple constraint settings, including peak-period, calendar, load-level, and volatility patterns. Additional held-out constraint-combination and downstream forecasting experiments further demonstrate that UPLOTS generalizes beyond the original peak-pattern setting and improves data augmentation under scarce real-data regimes. Our code and baselines are available at anonymous github repo: https://anonymous.4open.science/r/UPLOTS-6C36.

We show the standard basis of transformer hidden states already provides a training-free, architecture-general feature basis. Individual dimensions encode semantic content via their signs (+/-1) and confidence via their magnitudes, acting as independent binary registers; a feature is a subset of dimensions with a consistent sign pattern, read by counting sign agreements with no learned rotation. We validate this Bag of Dims framework across seven models spanning language (Qwen 3.5-4B, Gemma 3-4B, Mistral 7B, Qwen3-32B), vision (DINOv2, ViT-Base), and audio (AST). Signs alone carry predictive content: unit-magnitude sign patterns preserve 60-93% top-5 next-token accuracy through the LM head, and decoder-free Hamming scoring reaches 80-90% top-4096. From a single-token cache (one forward pass per token, no context, no labels), we detect 175 categories at AUC 0.97-0.99 by sign agreement; a trained probe adds only +0.018 AUC and converges to axis-aligned weights. These features are causally operative: they survive the K/V attention projections, trace to the FFN neuron coalitions that write them (random-weight controls never reproduce this), and flipping a feature's signs during the live forward pass suppresses its concept across four language models, magnitude-matched and concept-specific. Dimensions stay independent throughout (pairwise mutual information below 0.006 bits). The structure is not specific to language: the same per-dimension signs appear in self-supervised vision (DINOv2, 9/12 ImageNet superclasses), supervised vision (ViT-Base, 11/12), and audio (AST, 50/50 ESC-50 categories), so it reflects transformer training in general, not the language-modeling objective. The standard basis already suffices for feature reading at one forward pass, no optimization, no GPU-days. The open problem shifts from finding the right rotation to cataloging what each dimension encodes.

Adaptive Hamiltonian learning is central to calibrating and characterizing quantum devices. In an adaptive controller, choosing the next experiment is itself a computation. Bayesian design rules are recomputed after every posterior update, and that step can take seconds. Across hundreds of shots, those seconds become a significant wall-clock cost for adaptivity. We introduce SymQNet, an amortized reinforcement-learning approach for low-latency adaptive Hamiltonian learning. SymQNet learns a posterior-conditioned acquisition policy offline, then uses a fast policy forward pass online while retaining Bayesian posterior feedback. On transverse-field Ising benchmarks, SymQNet substantially reduces acquisition latency relative to bounded Fisher-information search and bounded two-step Bayesian active learning by disagreement (BALD). At five qubits, it reduces acquisition-only decision latency by $47.1\times$ and $72.6\times$ relative to these online baselines; at twelve qubits, full simulated steps take $1.02$ s for SymQNet versus $13.27$ s for bounded two-step BALD. Overall, we show that learned acquisition can make adaptive Hamiltonian learning practical for repeated low-latency workloads.

Modern deep learning models remain notoriously prone to overconfidence, limiting their reliability in high-stakes applications. Bayesian methods aim to counter this by learning a distribution over model parameters, and recent advances now make this feasible for large-scale architectures at costs comparable to AdamW. However, a challenge remains at test time: predictions must be averaged across many forward passes with weights sampled from the posterior, which is prohibitively expensive. Variance propagation offers an efficient alternative, computing layer-wise analytical approximations of uncertainty in a single forward pass. While such techniques are effective for MLPs, their extension to modern architectures remains challenging, due to increased depth and diversity of layer types. To fill this gap, we propose Calibrated Variance Propagation (CVP), which introduces a new propagation method for normalization layers, combines it with recent techniques for handling activation functions, and absorbs residual error through a light calibration step. CVP yields comparably accurate uncertainty estimates to MC sampling across transformers and CNNs, at a fraction of the cost. Against prior variance propagation work, CVP improves coverage at $0.5\%$ risk from $8.2\%$ to $14.6\%$ with BEiT-3 on Visual Reasoning (NLVR2) and from $2.6\%$ to $10.8\%$ with ViLT on VQAv2, with gains extending to convolutional architectures.

As large language models (LLMs) are increasingly deployed in real-world applications, alignment is no longer governed by a single universal notion of safety or helpfulness, but instead by provider- or application-specific model specifications. These specifications are typically long, structured, and frequently updated, yet existing alignment pipelines lack a systematic mechanism to operationalize them as training signals. In this paper, we propose specification-grounded alignment, a new alignment paradigm that treats provider-authored model specifications as the primary alignment target rather than abstract principles or static benchmarks. To instantiate this paradigm, we introduce SpecAlign, a framework that synthesizes alignment data directly from specification documents. SpecAlign combines structured rule annotation, controllable specification instantiation, and multi-agent adversarial data synthesis to generate fine-grained, boundary-aware preference pairs that capture both compliant behaviors and meaningful specification violations. Experiments across multiple model specifications and backbone models demonstrate that training with SpecAlign consistently improves rule compliance while preserving general capabilities and avoiding over-conservative behavior. These results suggest that grounding alignment in explicit model specifications enables rapid, precise, and scalable adaptation of LLM behavior to evolving policy requirements.

Controllable image-to-video (I2V) generation transforms a reference image into a coherent video guided by user-specified control signals. While precise control over camera motion, object motion, and lighting is essential for high-fidelity creation, existing methods often treat these factors independently. This overlooks the physical coupling among viewpoint, geometry, and illumination in dynamic scenes, leading to visual inconsistencies such as mismatched shadows and perspective drift under simultaneous changes. We present VidCRAFT3, a unified and flexible I2V framework that explicitly models cross-factor interactions among geometry, motion, and illumination, enabling both independent and joint control over camera motion, object motion, and lighting direction. Image2Cloud provides explicit 3D geometric priors for accurate camera motion control. ObjMotionNet encodes sparse object trajectories into multi-scale motion features to guide realistic object motion. A Spatial Triple-Attention Transformer integrates lighting direction through lighting cross-attention for consistent relighting. To address the scarcity of jointly annotated data, we construct the VideoLightingDirection (VLD) dataset with accurate per-frame lighting direction annotations, and introduce a three-stage progressive training strategy that enables robust learning without fully joint annotations. Extensive experiments demonstrate that VidCRAFT3 achieves state-of-the-art performance in control precision and visual coherence across diverse scenarios.

Time series forecasting plays a vital role in supporting decision-making across a wide range of critical applications, including energy, healthcare, and finance. Despite recent advances, forecasting accuracy remains limited due to the challenge of integrating historical numerical sequences with contextual features, which often comprise unstructured textual data. To address this challenge, we propose TokenCast, a large language model (LLM) driven framework that leverages language-based symbolic representations as a unified intermediary for context-aware time series forecasting. Specifically, TokenCast employs a discrete tokenizer to transform continuous numerical sequences into temporal tokens, enabling structural alignment with language-based inputs. To effectively bridge the semantic gap between modalities, both temporal and contextual tokens are embedded into a shared representation space via a pre-trained LLM, further optimized with generative objectives. Building upon this unified semantic space, the aligned LLM is subsequently fine-tuned in a supervised manner to predict future temporal tokens, which are then decoded back into the original numerical space. Extensive experiments on real-world datasets demonstrate the effectiveness of our framework and highlight its potential as a generative framework for context-aware time series forecasting. The code is available at https://github.com/Xiaoyu-Tao/TokenCast.

Large language models (LLMs) achieve strong performance on metaphor detection and interpretation tasks, yet it remains unclear what such behavioral success reveals about metaphor processing. We present a diagnostic analysis that examines the limits of behavioral evidence by probing three complementary dimensions: semantic attribute alignment, lexical invariance, and syntactic sensitivity. Using geometric probing, we assess whether model-generated interpretations align with reference semantic attributes; through context-varying substitution, we analyze the stability of lexical associations between metaphorical and literal expressions; and via controlled syntactic perturbations, we examine sensitivity in metaphor detection. Our analysis reveals that LLM-generated interpretations can exhibit semantic drift relative to reference attributes; stable lexical anchors persist across contextual conditions, potentially supporting conventional metaphors while biasing novel metaphors requiring contextual integration; and detection performance is sensitive to syntactic irregularities. These findings suggest that strong behavioral performance may reflect heterogeneous underlying signals, highlighting the need for caution when interpreting metaphor benchmarks as evidence of robust, integrated semantic understanding.

Any piece of knowledge is usually expressed in one or a handful of natural languages on the web or in any large corpus. Large Language Models (LLMs) act as a bridge by acquiring knowledge from a source language and making it accessible when queried using target languages. A cross-lingual gap is a drop in accuracy incurred when querying knowledge in a target language rather than the source language. Existing research focused on modeling or training failures leading to cross-lingual gaps. In this work, we take an alternative view to characterize the nature of cross-lingual error, and hypothesize that the variance of responses in the target language is a key cause of this gap. For the first time, we formalize the cross-lingual gap in terms of biased and unbiased errors. We empirically validate our hypothesis through multiple inference-time interventions that control variance and reduce the cross-lingual gap. We demonstrate a few test-time ensemble methods that reduce response variance, and thereby improve source-target transfer scores by up to 12 absolute points yielding relative gains of 8% to over 50% across various LLMs.

Most existing time series classification methods adopt a discriminative paradigm that maps input sequences directly to one-hot encoded class labels. While effective, this paradigm struggles to incorporate contextual features and fails to capture semantic relationships among classes. To address these limitations, we propose InstructTime, a novel framework that reformulates time series classification as a multimodal generative task. Specifically, continuous numerical sequences, contextual textual features, and task instructions are treated as multimodal inputs, while class labels are generated as textual outputs by tuned language models. To bridge the modality gap, InstructTime introduces a time series discretization module that converts continuous sequences into discrete temporal tokens, together with an alignment projection layer and a generative self-supervised pre-training strategy to enhance cross-modal representation alignment. Building upon this framework, we further propose InstructTime++, which extends InstructTime by incorporating implicit feature modeling to compensate for the limited inductive bias of language models. InstructTime++ leverages specialized toolkits to mine informative implicit patterns from raw time series and contextual inputs, including statistical feature extraction and vision-language-based image captioning, and translates them into textual descriptions for seamless integration. Extensive experiments on multiple benchmark datasets demonstrate the superior performance of InstructTime++.

Counterfactual story retelling exposes LLM shortcomings in constrained narrative solution spaces where they can no longer rely on recalling memorised training data. Ground-truth-based post-training, such as SFT, fails to teach LLMs how to generate logical and rational narrative events. In this paper, we introduce Retell, Reward, Repeat (RRR), an RL-based pipeline synthesising Structuralist Narratology with scalar narrativity to teach storytelling structure. We extend the TimeTravel dataset with human-annotated stages of narrative equilibrium to evaluate reward models. By using d-RLAIF, RRR derives training signals from the narrativity of textual features without the need for reference outputs. Evaluations demonstrate that RRR-trained LLMs outperform few-shot and SFT baselines in logic, rationality, and completeness, with output quality additionally validated by blind human preference. Relying on a small, query-only dataset, RRR provides a linguistically grounded, cost-effective post-training mechanism for storytelling--a domain currently lacking effective post-training methods. RRR highlights the continued relevance of integrating established linguistic theories into contemporary NLP.

For generative AI agents to partner effectively with human users, the ability to accurately predict human intent is critical. But this ability to collaborate remains limited by a critical deficit: an inability to model common ground. We present a referential communication experiment with a factorial design involving director-matcher pairs (human-human, human-AI, AI-human, and AI-AI) that interact with multiple turns in repeated rounds to match pictures of objects not associated with any obvious lexicalized labels. We show that LVLMs cannot interactively generate and resolve referring expressions in a way that enables smooth communication, a crucial skill that underlies human language use. We release our corpus of 356 dialogues (89 pairs over 4 rounds each) along with the online pipeline for data collection and the tools for analyzing accuracy, efficiency, and lexical overlap.

Scaling training data and model parameters has long driven progress in large language models (LLMs), but this paradigm is increasingly constrained by the scarcity of high-quality data and diminishing returns from rising computational costs. As a result, recent work is increasing the focus on continual learning from real-world deployment, where user interaction logs provide a rich source of authentic human feedback and procedural knowledge. However, learning from user logs is challenging due to their unstructured and noisy nature. Vanilla LLM systems often struggle to distinguish useful feedback signals from noisy user behavior, and the disparity between user log collection and model optimization (e.g., the off-policy optimization problem) further strengthens the problem. To this end, we propose UNO (User log-driveN Optimization), a unified framework for improving LLM systems (LLMsys) with user logs. UNO first distills logs into semi-structured rules and preference pairs, then employs query-and-feedback-driven clustering to manage data heterogeneity, and finally quantifies the cognitive gap between the model's prior knowledge and the log data. This assessment guides the LLMsys to adaptively filter out noisy feedback and construct different modules for primary and reflective experiences extracted from user logs, thereby improving future responses. Extensive experiments show that UNO achieves state-of-the-art effectiveness and efficiency, significantly outperforming Retrieval Augmented Generation (RAG) and memory-based baselines. We have open-sourced our code at https://github.com/bebr2/UNO .

Applications of narrative theories using large language models (LLMs) deliver promising methods in automatic story generation and understanding tasks. Our survey examines how natural language processing (NLP) research uses LLM methods to engage with diverse concepts from narrative studies. We use established distinctions from narratology to categorise ongoing efforts and discover the following: \redtext{(a) narrative texts come from diverse sources beyond just literature, (b) theoretical synthesis and validation are potential outcomes, (c) generation tasks lag behind understanding in several ways: theoretical application, post-training methods, exploring non-fiction narratives and addressing narrative levels beyond fabula and discourse.} For future directions, instead of the pursuit of a single, generalised benchmark for `narrative quality', we believe that progress can benefit from efforts that focus on the following: defining and improving theory-based metrics for individual narrative attributes; continue conducting large-scale, theory-driven literary/social/cultural analysis; generating narratives in situated contexts; and continuing experiments where outputs can be used to validate or refine narrative theories. This work provides a contextual foundation for more systematic and theoretically informed narrative research in NLP by providing an overview to ongoing research efforts and the broader narrative studies landscape.

Autoregressive long video generation often adopts bounded-memory streaming for efficiency, typically combining local windows for short-term continuity with static early-frame sinks as long-range anchors. However, this fixed allocation keeps early frames cached even when the current visual state has substantially diverged from them, while discarding potentially more relevant intermediate history. As a result, the retained long-range context may become less adaptive and bias generation toward outdated cues; in severe cases, RoPE-induced phase re-alignment can homogenize inter-head attention and cause sink collapse, where content regresses toward sink frames. We propose DySink, a retrieval-based framework that maintains a compact memory bank and selects visually relevant historical frames as dynamic frame sinks. DySink couples adaptive retrieval with a sink anomaly gate, which detects excessive inter-head consensus over retrieved context and suppresses collapse-prone context. Experiments on minute-long videos show that DySink consistently improves temporal quality over strong baselines while also achieving higher dynamic degree, enabling coherent and more natural long-horizon visual evolution. The code and model weights are released at https://github.com/yebo0216best/DySink.

Cinematic video depicts multiple subjects acting or interacting at specific moments, captured with deliberate camera movement, and stitched together by shot transitions. Together, these elements demand a level of fine-grained control beyond current text-to-video models. Existing work addresses each axis in isolation: multi-subject personalization, temporal control, multi-shot synthesis, or camera control; no prior framework jointly integrates all four. We present CineOrchestra, a unified video diffusion model that controls subjects, events, cameras, and shot transitions simultaneously. Our key insight is that these heterogeneous cinematic elements share a fundamental structure: each is an entity acting over a specific temporal interval, which can therefore all be expressed through one shared structure of entity-centric conditioning primitives, augmented with reference images for visual entities. This formulation reduces the architectural challenge to a single positional encoding problem, which we solve with two parameter-free coordinated rotary embeddings: (a) an interval-sampled temporal RoPE that yields consistent attention behavior across events of dramatically varying duration, and (b) a 2D entity-temporal cross-attention RoPE that disambiguates per-entity conditions and routes each to its corresponding spatiotemporal region. On two new benchmarks, CineOrchestra outperforms six per-axis specialists on dense caption following and shot-transition timing, with consistent gains in a pairwise user study and component ablations. Project page: https://snap-research.github.io/CineOrchestra

Two recent studies (Jones et al. (2026); Zeng et al. (2026)) reach apparently contradictory conclusions about whether LVLMs can coordinate on efficient referring expressions. We control for task differences between the studies while directly comparing their prompting styles. We replicate the finding that models can coordinate efficient referring expressions when explicitly prompted to do so, suggesting that other task differences are not responsible for divergent results. However, we also find that the same models fail to infer the need for communicative efficiency from a more implicit prompt, highlighting critical differences between how humans and AI systems communicate.

Pathological images are inherently multi-scale, requiring pathologists to integrate evidence from global tissue architecture at low magnification to cellular morphology at higher magnification for accurate diagnosis. While existing pathological datasets for vision-language model (VLM) include various scales, they often lack an explicit cross-scale reasoning objective. This limitation prevents VLMs from capturing essential cross-scale representations and learning evidence-based reasoning. To bridge this gap, we introduce the first cross-scale training and evaluation paradigm that formulates pathology interpretation as multi-magnification reasoning. However, creating such a task reveals a critical challenge: multi-image visual question answering (VQA) is prone to text-only shortcuts, which allow models to guess answers using magnification-dependent artifacts rather than visual evidence. To address this, we propose a leakage-aware curation pipeline that combines adversarial text-only screening with constraint-guided question design. Using this pipeline, we construct Scale-VQA, a high-quality benchmark with 4,685 multiple-choice questions grounded in 2,537 pathology images across multiple magnification levels. Finally, we present ScaleReasoner-R1, a model trained via reinforcement learning to optimize performance on the cross-scale VQA task. ScaleReasoner-R1 achieves state-of-the-art performance on our cross-scale reasoning benchmark and generalizes to SOTA performance on established single-scale benchmarks. Findings suggest that even the limited cross-scale supervision can significantly improve pathological understanding. The code and demos will be open-sourced.

We introduce Cosmos 3, a family of omnimodal world models designed to jointly process and generate language, image, video, audio, and action sequences within a unified mixture-of-transformers architecture. By supporting highly flexible input-output configurations, Cosmos 3 seamlessly unifies critical modalities for Physical AI -- effectively subsuming vision-language models, video generators, world simulators, and world-action models into a single framework. Our evaluation demonstrates that Cosmos 3 establishes a new state-of-the-art across a diverse suite of understanding and generation tasks, demonstrating omnimodal world models as scalable, general-purpose backbones for embodied agents. Our post-trained Cosmos 3 models were ranked as the best open-source Text-to-Image and Image-to-Video models by Artificial Analysis, and the best policy model by RoboArena at the time the technical report was written. To accelerate open research and deployment in Physical AI, we make our code, model checkpoints, curated synthetic datasets, and evaluation benchmark available under the Linux Foundation's OpenMDW-1.1 License at https://github.com/nvidia/cosmos and https://huggingface.co/collections/nvidia/cosmos3. The project website is available at https://research.nvidia.com/labs/cosmos-lab/cosmos3.

When an AI assistant remembers that Sarah is a single mother working two jobs, does it interpret her stress differently than if she were a wealthy executive? As personalized AI systems increasingly incorporate long-term user memory, understanding how this memory shapes emotional reasoning is critical. We investigate how user memory affects emotional intelligence in large language models (LLMs) by evaluating 15 models on human-validated emotional intelligence tests. We find that identical scenarios paired with different user profiles produce systematically divergent emotional interpretations. Across validated user-independent emotional scenarios and diverse user profiles, systematic biases emerged in several high-performing LLMs where advantaged profiles received more accurate emotional interpretations. Moreover, LLMs demonstrate significant disparities across demographic factors in emotion reasoning and supportive recommendations tasks, indicating that personalization mechanisms can embed social hierarchies into models' emotional reasoning. These results highlight a key challenge for memory-enhanced AI: systems designed for personalization may reinforce social inequalities. To mitigate these disparities, we curate a general-purpose preference dataset designed to reduce demographic profiles' influence on emotional understanding.

Robust lie detectors for language models could enable powerful techniques for auditing, monitoring, and post-hoc investigation of model behaviour, but evaluating them requires testbeds where models verifiably believe the opposite of what they say. We show that existing trained model organisms often fail this requirement, leaving prior positive and negative detection results difficult to interpret. We address this with 13 reasoning model organisms whose hidden beliefs are verified in chain-of-thought and shown to generalise to held-out tasks, alongside Varied Deception, a prompted-lying testbed covering a broad range of lie-inducing motivations. On these testbeds we evaluate four detectors: a chain-of-thought judge, a logprob classifier, and two activation probes, including Did-You-Lie (DYL), a new method for training follow-up probes. On prompted lying, across 31 open-weight models spanning 2B to 1T parameters, all four detectors show positive scaling with model capability. However, every activation- and logprob-based detector drops sharply on our trained model organisms, with DYL retaining the most signal; only the chain-of-thought judge remains strong, with 0.82 balanced accuracy, partly as an artefact of our verification process favouring CoT-readable beliefs. Current lie detectors therefore cannot support high-confidence claims about model beliefs, and we suggest research directions that may address some of their current limitations. We release our datasets, model organisms, and trained detectors.

The increasing use of artificial intelligence (AI) in news production raises important questions about how audiences perceive and respond to AI-generated journalism. This preregistered survey experiment (N = 599, German-speaking Switzerland) examines (i) perceptions of article quality (measured as credibility, readability, and expertise) across news excerpts that were human-written, AI-assisted, or fully AI-generated, and (ii) self-reported intentions to engage following disclosure of AI involvement. Participants rated two short news excerpts before learning how they had been produced. Articles across all conditions were evaluated similarly in perceived quality. After disclosure, participants in the AI-assisted and AI-generated conditions reported a higher willingness to continue reading their assigned articles compared to the control group, but future willingness to read AI-generated news did not differ across conditions. Overall, the findings suggest that readers assess AI-generated and human-written news comparably in quality, while disclosure of AI use can momentarily increase curiosity or interest without yet changing longer-term reading intentions.

Adversarial robustness of deep autoencoders (AEs) has received less attention than that of discriminative models, although their compressed latent representations induce ill-conditioned mappings that can amplify small input perturbations and destabilize reconstructions. Existing white-box attacks for AEs, which optimize norm-bounded adversarial perturbations to maximize reconstruction damage, often converge to suboptimal perturbations, thereby potentially overstating AE robustness. We show that this limitation is linked to vanishing adversarial loss gradients during backpropagation through ill-conditioned layers, associated with near-zero singular values in their intermediate weight matrices. To address this, we propose GRILL (Gradient Signal Restoration in Ill-Conditioned Layers), a framework designed to mitigate gradient degradation and improve the reliability of adversarial robustness evaluation in encoder-decoder architectures. GRILL is designed to mitigate adversarial gradient degradation during optimization, enabling attacks to better approximate high-distortion perturbations under fixed norm constraints. Through extensive experiments across multiple AE architectures, under both sample-specific and universal attacks, as well as standard and adaptive attack settings, we show that GRILL significantly increases attack effectiveness, thereby exposing vulnerabilities hidden by existing attack limitations. Beyond AEs, we provide preliminary evidence that modern multimodal encoder-decoder architectures exhibit similar vulnerabilities.

AI-Generated (AIG) content has become increasingly widespread by recent advances in generative models and the easy-to-use tools that have significantly lowered the technical barriers for producing highly realistic audio, images, and videos through simple natural language prompts. In response, platforms are adopting provable provenance with platforms recommending AIG to be self-disclosed and signaled to users. However, these indicators may be often missed, especially when they rely solely on visual cues and make them ineffective to users with different sensory abilities. To address the gap, we conducted semi-structured interviews (N=28) with 15 sighted and 13 BLV participants to examine their interaction with AIG content through self-disclosed AI indicators. Our findings reveal diverse mental models and practices, highlighting different strengths and weaknesses of content-based (e.g., title, description) and menu-aided (e.g., AI labels) indicators. While sighted participants leveraged visual and audio cues, BLV participants primarily relied on audio and existing assistive tools, limiting their ability to identify AIG. Across both groups, they frequently overlooked menu-aided indicators deployed by platforms and rather interacted with content-based indicators such as title and comments. We uncovered usability challenges stemming from inconsistent indicator placement, unclear metadata, and cognitive overload. These issues were especially critical for BLV individuals due to the insufficient accessibility of interface elements. We provide practical recommendations and design implications for future AIG indicators across several dimensions.

Current unlearning methods for LLMs optimize on the private information they seek to remove by incorporating it into their fine-tuning data. We argue this not only risks reinforcing exposure to sensitive data, but also fundamentally contradicts the principle of minimizing its use. As a remedy, we propose a novel unlearning method-Partial Model Collapse (PMC), which does not require unlearning targets in the unlearning objective. Our approach is inspired by recent observations that training generative models on their own generations leads to distribution collapse, effectively removing information from model outputs. Our central insight is that model collapse can be leveraged for machine unlearning by deliberately triggering it for data we aim to remove. We theoretically analyze that our approach converges to the desired outcome, i.e. the model unlearns the data targeted for removal. We empirically demonstrate that PMC overcomes four key limitations of existing unlearning methods that explicitly optimize on unlearning targets, and more effectively removes private information from model outputs while preserving general model utility. Overall, our contributions represent an important step toward more comprehensive unlearning that better aligns with real-world privacy constraints. Code available at https://www.cs.cit.tum.de/daml/partial-model-collapse/.

While prior research on text-to-image generation has predominantly focused on biases in human depictions, demographic bias in generated objects remains relatively underexplored. We introduce SODA (Stereotyped Object Diagnostic Audit), a novel framework for systematically measuring these biases through automated attribute discovery and three standardized metrics: Base vs. Demographic Divergence (BDS), Cross-Demographic Disparity (CDS), and Visual Attribute Concentration (VAC). Applying SODA to 8,000 images across five state-of-the-art models and eight object categories (e.g., cars), we find that "neutral" prompts produce outputs most visually similar to middle-aged and White people, suggesting these groups are implicitly over-represented in model defaults. Furthermore, demographic cues trigger highly skewed stereotypical outputs: 26.6% of object-model-demographic combinations produce results where all 20 generated images share the exact same attribute value (e.g., rose gold laptops for women). Finally, prompt-level debiasing reduces inter-group disparity but paradoxically collapses within-group diversity, replacing one stereotype with another. SODA offers a practical pipeline for making these implicit associations measurable, serving as a step toward more responsible AI development.

Multimodal Large Language Models (MLLMs) are increasingly deployed in stateless systems, such as autonomous driving and robotics. This paper investigates a novel threat: Semantic-Aware Hijacking. We explore the feasibility of hijacking multiple stateless decisions simultaneously using a single universal perturbation. We introduce the Semantic-Aware Universal Perturbation (SAUP), which acts as a semantic router, "actively" perceiving input semantics and routing them to distinct, attacker-defined targets. To achieve this, we conduct theoretical and empirical analysis on the geometric properties in the latent space. Guided by these insights, we propose the Semantic-Oriented (SORT) optimization strategy and annotate a new dataset with fine-grained semantics to evaluate performance. Extensive experiments on three representative MLLMs demonstrate the fundamental feasibility of this attack, achieving a 66% attack success rate over five targets using a single frame against Qwen.

Jailbreak attacks expose a persistent failure mode in safety-aligned LLMs: models can be pushed into harmful behavior, but the internal representations enabling this shift remain poorly localized. Recent mechanistic safety studies often explain such behavior through broad representational objects, including global refusal directions, activation steering vectors, and refusal-related SAE features. We instead ask whether jailbreak vulnerability can be traced to finer-grained, prompt-conditioned SAE feature subgroups. We introduce a token-driven mechanistic pipeline that decomposes the residual stream of Gemma-2-2B into Sparse Autoencoder (SAE) features and identifies feature subgroups associated with unsafe behavior. Using single-category unsafe examples from BeaverTails to reduce cross-category interference, we extract harmful concepts from adversarial responses and align them with concept-relevant prompt tokens through subspace similarity. We then apply three feature-grouping strategies: cluster-based, hierarchical-linkage, and single-token-driven, to identify SAE feature subgroups across all 26 layers. Finally, we amplify the top features in each subgroup and evaluate the resulting generations with a standardized harmfulness judge. Single-token-driven grouping achieves harmfulness comparable to full cluster-based grouping, showing that individual harmful prompt tokens are sufficient to localize vulnerability-relevant SAE feature subgroups without relying on broader cluster-level aggregation. These subgroups appear across early and mid-to-late layers, with stronger concentration in mid-to-late layers, where targeted steering exposes specific model vulnerabilities. Overall, our results suggest that jailbreak susceptibility can be traced to sparse, token-localized SAE feature subgroups, complementing prior accounts based on broad adversarial, refusal, or steering directions.

Structured data is well handled by gradient-boosted decision trees (GBDT), which are usually trained on vertically partitioned features across mutually distrustful parties. High speed and interpretability make GBDTs popular in finance and healthcare, where neural networks may fall short. Enabling secure computation for GBDTs poses unique challenges, requiring secure record alignment for comparison. Relying on private set intersection (PSI) is a de facto approach. Mistaking PSI for a safety measure actually exposes which record identifiers (IDs) are shared between the datasets. Although circuit-PSI could help, it is costly for generic uses. New ideas are needed to efficiently train in a "dark forest". Aiming to hide the IDs, we initiate the study of anonymous GBDT training on split data held by two parties. Dual circuit-PSI in our design lets the parties alternate as receiver to run pick-then-sum over local features. Via oblivious programmable pseudorandom functions, we propagate circuit-PSI outputs as shared state across runs. Avoiding universal alignment, we resolve the neglected dilemma that ID hiding incurs a cost that scales with domain size. Next, we halve the cost of ciphertext packing used to convert single-instruction multiple-data homomorphic encryption from (ring) learning with errors in prior secure GBDT (Usenix Security' 23) and related secure machine-learning computations. Comparative experiments show our protocol remains competitive with leaky approaches in efficiency. Enabling ID-hiding aggregation, our techniques can extend to other vertically partitioned analytics.

Agent-interoperability protocols such as A2A and MCP standardize what agents say to one another but assume address-based transport. Whether over HTTP(S) or a content-protecting binding such as MLS-based SLIM, these transports protect message content yet leave the communication graph exposed: which agent contacts which, when, and how often. In agent systems this graph is more consequential than a privacy framing suggests. Endpoints are capability-labeled, workflows are structured and chained, and interactions are coupled to actions, so an observer recovers more than past relationships: it can recognize a recurring pending workflow from its opening and, at machine speed, act on it before it completes. The threat is one of workflow integrity, not privacy alone. We give a threat model for the communication graph and locate what makes its metadata distinctively consequential: not stronger fingerprinting but exposure across independent trust domains, coupled to autonomous action. We define transport- and bootstrap-layer privacy properties, give them an indistinguishability-game semantics, evaluate transports, and give an A2A case study where a metadata-protecting binding surfaces its implicit identity assumptions. On a corpus of real multi-agent A2A traffic from the official reference agents, on a live A2A binding, and with a generative model as a controlled instrument, a label-blind classifier recovers a task's class from passive metadata at 6x chance, and from only its opening; a defense-aware adversary does not overturn this, and only the full set of properties drives recovery toward chance. Acting on the leak is distinct from recoverability: under a fixed budget an adversary captures 0.63 of a clairvoyant attacker's advantage on the corpus (0.41 from a workflow's opening), governed by top-ranked precision rather than overall accuracy, so integrity and privacy come apart under defense.

Targeted interventions on language models, such as unlearning or model editing, aim to modify specific information, but their effects often propagate to related, unintended areas (e.g., removing virology content may degrade performance on allergies); these side-effects are commonly referred to as the ripple effect. We introduce RippleBench-Maker, an automatic pipeline that retrieves semantic neighbors of any source concept from a knowledge repository and generates multiple-choice questions at varying semantic distances. We instantiate this framework using WikiRAG, an open-source RAG system over English Wikipedia, to construct RippleBench-WMDP-Bio (584 seed topics, 352,961 questions), and evaluate eight unlearning methods on Llama3-8B-Instruct. All eight exhibit accuracy drops that are largest near the unlearned target and decay with semantic distance, each with a distinct propagation profile. We replicate these findings across Mistral-7B, Zephyr-7B, and Yi-34B; cross-model delta curves are nearly identical, suggesting ripple effects are a property of the unlearning method rather than the base model. We validate all major pipeline stages using a four-experiment Mechanical Turk study (5,200+ responses, 61 workers). We release all code, data, and infrastructure.

Large language models in Agentic AI systems consume tool schemas and execution results and emit tool invocations as structured data. The default language for that exchange, JSON, was designed for application-to-application interchange rather than token efficiency, so its structural elements impose substantial token overhead. Recent work proposes token-optimized alternatives such as TOON (Token-Oriented Object Notation) and TRON (Token Reduced Object Notation) as more compact replacements, but these formats have been evaluated only on isolated comprehension or generation tasks. Whether their token reductions hold inside end-to-end agentic loops therefore remains an open question. We evaluate TOON and TRON on four agentic benchmarks (BFCL, MCPToolBenchPP, MCP-Universe, StableToolBench) and five open-weight LLMs, decoupling input compression from output compression to measure comprehension and generation independently. TRON reduces tokens by up to 27% with accuracy within 14pp of the JSON baseline. TOON achieves up to 18% reduction at a similar 9pp accuracy cost, but additionally cascades on multi-turn parsing failures and collapses parallel tool-call output for most models. The code is available at: https://github.com/lkutschka/notation-matters

Large language model agents are increasingly integrated into map services. Since map services are embedded in everyday-life scenarios rather than professional task settings, users often express their needs informally, resulting in underspecified queries with many unspoken needs, namely, implicit decision factors that are critical for user satisfaction. Although clarification is an effective way to mitigate this issue, it increases user burden in daily interaction, and a capable agent should first proactively recover such factors from available information sources. However, evaluating this ability is challenging. The first challenge is to determine which implicit decision factors are suitable for evaluation. A factor is evaluable only if it affects user acceptance and can be recovered from information available to the agent before it responds. Second, user satisfaction cannot be reliably represented by a single reference answer, requiring a benchmark that converts satisfaction-relevant factors into objective and quantifiable evaluation targets. To address these challenges, we propose a restore-identify-filter framework that reconstructs complete user needs from behavior-chain evidence, identifies implicit decision factors, and retains only those supported by pre-query evidence. Building on this methodology, we construct MapSatisfyBench from large-scale, real-world anonymized user data and annotate ground truth from five dimensions and enables full-chain evaluation of satisfaction-aware map agents. Experiments show that current agents generally perform well on explicit task completion, but remain limited in satisfying implicit decision factors and proactively acquiring the evidence needed for satisfaction-aware decisions. These findings establish MapSatisfyBench as a benchmark for shifting map-agent evaluation from task completion toward satisfaction-aware spatial decision making.

AI agents are moving from advisors to actors, booking travel, planning menus, and running procurement on behalf of users. Existing benchmarks for AI and animal welfare evaluate model text responses to question-answer prompts, leaving open whether the welfare reasoning surfaced in those responses transfers to agentic deployment where the model must take actions with tools. We introduce TAC (Travel Agent Compassion), the first agentic benchmark measuring whether AI agents avoid options involving animal exploitation when acting on behalf of users. TAC presents an AI agent with twelve hand-authored travel booking scenarios across six categories of animal exploitation, augmented to forty-eight samples to control for price, rating, and position confounds. We evaluate seven frontier models from four labs. Every model scores below the chance level of sixty-four percent, with the best performer (Claude Opus 4.7) at fifty-three percent. A single welfare-aware sentence in the system prompt yields gains of forty-seven to sixty-three percentage points in Claude and GPT-5.5, twenty-six points in GPT-5.2, and under twelve points in DeepSeek and Gemini. An auxiliary Inspect Scout audit of 288 base-condition transcripts from the top two performers, using Gemini 2.5 Flash Lite as judge, flags zero transcripts for evaluation awareness, suggesting the below-chance rates do not stem from the models recognising the evaluation. We discuss implications for category-level variation across cultural domains, the limits of text-response welfare benchmarks, and the EU General-Purpose AI Code of Practice systemic risk framework.

As high-quality public web corpora become increasingly exhausted, clean long-context documents have become a scarce and expensive source of training data for large language models (LLMs). Existing long-context corpora are often proprietary and costly to acquire, synthetically generated, or concentrated in narrow domains such as programming. We introduce the Stanford EDGAR Filings Dataset (SEFD), an open reconstruction of SEC filings into layout-faithful MultiMarkdown for financial language modeling and evaluation. SEFD makes audited financial statements, risk disclosures, ownership reports, accounting notes, and market-moving event filings usable as long-context pretraining data and as a basis for financial reasoning, forecasting, compliance, and document understanding. The resulting corpus is token-efficient, model-ready, and has less than 0.1% overlap with Common Crawl-derived corpora. We release SEFD-v1, a 152B-token initial public snapshot, and provide corpus-level analyses of a larger 18.5M-filing archive estimated at 550B tokens. We further introduce two SEFD-derived benchmarks: EDGAR-Forecast, which evaluates filing-grounded numerical forecasting after model knowledge cutoffs, and EDGAR-OCR, which evaluates transcription of complex financial tables.

In real-world applications, a machine learning model is required to handle an open-set recognition (OSR), where unknown classes appear during the inference, in addition to a domain shift, where the data distribution differs between the training and inference phases. Domain generalization (DG) aims to handle the domain shift situation where the target domain of the inference phase is inaccessible during the model training. Open domain generalization (ODG) considers DG and OSR. Domain-augmented meta-learning (DAML) is a method targeting ODG; however, it has a complicated learning process. By contrast, although various DG methods have been proposed, they have not been evaluated in ODG situations. In this study, we comprehensively evaluate the existing DG methods in ODG and show that the two simple DG methods, CORrelation ALignment (CORAL) and maximum mean discrepancy (MMD), are competitive with DAML in several cases. In addition, we propose simple extensions of CORAL and MMD by introducing the techniques used in DAML, such as ensemble learning and Dirichlet mixup data augmentation. The experimental evaluation demonstrates that the extended CORAL and MMD can perform comparably to DAML with lower computational costs. This suggests that the simple DG methods and their simple extensions are strong baselines for ODG.

We present UniTalk, a novel dataset emphasizing challenging scenarios to enhance model generalization for the task of active speaker detection (ASD). Previously established benchmarks such as AVA predominantly comprise old movies and thus exhibit significant domain gaps with real-world video. In contrast, UniTalk covers diverse video types reflecting challenging real-world conditions, including underrepresented languages, noisy backgrounds, and crowded scenes, while being on par with AVA in scale. Extensive evaluations reveal that ASD remains unsolved under realistic conditions: state-of-the-art models near-perfect on AVA fail to reach saturation on UniTalk. Conversely, models trained on UniTalk generalize better to modern in-the-wild datasets including Talkies and ASW. UniTalk thus establishes a new benchmark for ASD, providing researchers with a valuable resource for developing and evaluating versatile and resilient models.

Large language models (LLMs) are increasingly applied to symbolic mathematics, yet existing evaluations often conflate pattern memorization with genuine reasoning. To address this gap, we present ASyMOB, a high-resolution dataset of 35,368 validated symbolic math problems spanning integration, limits, differential equations, series, and hypergeometrics. Unlike prior benchmarks, ASyMOB systematically perturbs each seed problem using symbolic, numeric, and equivalence-preserving transformations, enabling a fine-grained assessment of generalization. Our evaluation reveals three key findings: (1) most models' performance collapses under minor perturbations, while top systems exhibit an apparent regime shift in robustness; (2) integrated code tools stabilize performance, particularly for weaker models; and (3) we identify examples where Computer Algebra Systems (CAS) fail while LLMs succeed, as well as problems solved only via a hybrid LLM-CAS approach, highlighting a promising integration frontier. ASyMOB serves as a principled diagnostic tool for measuring and accelerating progress toward building verifiable, trustworthy AI for scientific discovery.

Model merging techniques aim to integrate the abilities of multiple models into a single model. Most model merging techniques have hyperparameters, and their setting affects the performance of the merged model. Because several existing works show that tuning hyperparameters in model merging can enhance the merging outcome, developing hyperparameter optimization algorithms for model merging is a promising direction. However, its optimization process is computationally expensive, particularly in merging LLMs. In this work, we develop surrogate benchmarks for optimization of the merging hyperparameters to realize algorithm development and performance comparison at low cost. We define two search spaces and collect data samples to construct surrogate models to predict the performance of a merged model from a hyperparameter. We demonstrate that our benchmarks can predict the performance of merged models well and simulate optimization algorithm behaviors.

Adapting general-domain retrievers to scientific domains is challenging due to the scarcity of large-scale domain-specific relevance annotations and the substantial mismatch in vocabulary and information needs. Recent approaches address these issues through two independent directions that leverage large language models (LLMs): (1) generating synthetic queries for fine-tuning, and (2) generating auxiliary contexts to support relevance matching. However, both directions overlook the diverse academic concepts embedded within scientific documents, often producing redundant or conceptually narrow queries and contexts. To address this limitation, we introduce an academic concept index, which extracts key concepts from papers and organizes them guided by an academic taxonomy. This structured index serves as a foundation for improving both directions. First, we enhance the synthetic query generation with concept coverage-based generation (CCQGen), which adaptively conditions LLMs on uncovered concepts to generate complementary queries with broader concept coverage. Second, we strengthen the context augmentation with concept-focused auxiliary contexts (CCExpand), which leverages a set of document snippets that serve as concise responses to the concept-aware CCQGen queries. Extensive experiments show that incorporating the academic concept index into both query generation and context augmentation leads to higher-quality queries, better conceptual alignment, and improved retrieval performance.

Large language models (LLMs) have shown growing promise in biomedical research, particularly for knowledge-driven interpretation tasks. However, their ability to reliably reason from gene-level knowledge to functional understanding, a core requirement for knowledge-enhanced cell atlas interpretation, remains largely underexplored. To address this gap, we introduce SciHorizon-GENE, a large-scale gene-centric benchmark constructed from authoritative biological databases. The benchmark integrates curated knowledge for over 190K human genes and comprises more than 540K questions covering diverse gene-to-function reasoning scenarios relevant to cell type annotation, functional interpretation, and mechanism-oriented analysis. Motivated by behavioral patterns observed in preliminary examinations, SciHorizon-GENE evaluates LLMs along four biologically critical perspectives: research attention sensitivity, hallucination tendency, answer completeness, and literature influence, explicitly targeting failure modes that limit the safe adoption of LLMs in biological interpretation pipelines. We systematically evaluate a wide range of state-of-the-art general-purpose and biomedical LLMs, revealing substantial heterogeneity in gene-level reasoning capabilities and persistent challenges in generating faithful, complete, and literature-grounded functional interpretations. Our benchmark establishes a systematic foundation for analyzing LLM behavior at the gene scale and offers insights for model selection and development, with direct relevance to knowledge-enhanced biological interpretation.

Speech-aware large language models (LLMs) can accept speech inputs, yet their training objectives largely emphasize linguistic content or specific fields such as emotions or the speaker's gender, leaving it unclear whether they encode speaker identity. First, we propose a model-agnostic scoring protocol that produces continuous verification scores for both API-only and open-weight models, using confidence scores or log-likelihood ratios from the Yes/No token probabilities. Using this protocol, we benchmark recent speech-aware LLMs and observe weak speaker discrimination (EERs above 20% on VoxCeleb1). Second, we introduce a lightweight augmentation that equips an LLM with ASV capability by injecting frozen ECAPA-TDNN speaker embeddings through a learned projection and training only LoRA adapters. On TinyLLaMA-1.1B, the resulting ECAPA-LLM achieves 1.03% EER on VoxCeleb1-E, approaching a dedicated speaker verification system while preserving a natural-language interface.

Web agents automate browser tasks, ranging from simple form completion to complex workflows like ordering groceries. While current benchmarks evaluate general-purpose performance~(e.g., WebArena) or safety against malicious actions~(e.g., SafeArena), no existing framework assesses an agent's ability to successfully execute user-facing website security and privacy tasks, such as managing cookie preferences, configuring privacy-sensitive account settings, or revoking inactive sessions. To address this gap, we introduce WebSP-Eval, an evaluation framework for measuring web agent performance on website security and privacy tasks. WebSP-Eval comprises 1) a manually crafted task dataset of 200 task instances across 28 websites; 2) a robust agentic system supporting account and initial state management across runs using a custom Google Chrome extension; and 3) an automated evaluator. We evaluate a total of 8 web agent instantiations using state-of-the-art multimodal large language models, conducting a fine-grained analysis across websites, task categories, and UI elements. Our evaluation reveals that current models suffer from limited autonomous exploration capabilities to reliably solve website security and privacy tasks, and struggle with specific task categories and websites. Crucially, we identify stateful UI elements are a primary reason for agent failure, with toggles causing more than 45% task failure across many models.

Instruction-tuned LLMs can annotate thousands of instances at low cost. This raises two questions for active learning (AL): can LLM labels replace human labels within the AL loop, and does AL remain necessary when entire corpora can be cheaply labeled? We investigate both on a new dataset of 277,902 German political TikTok comments (25,974 LLM-labeled, 5,000 human-annotated), comparing LLM and human annotation across seven conditions, four encoders, and 10 random seeds. Under a two-question interface that mirrors the human annotation task, LLM annotation at scale outperforms human-supervised classifiers at roughly one-tenth the cost (\$28 for GPT-5.2 Batch API vs. \$316 for Prolific). The advantage holds for both a closed-source (GPT-5.2) and an open-weight (Qwen3.5-122B-10B) LLM, is robust under soft-label evaluation, and is unlocked specifically by the two-question decomposition; a holistic single-prompt baseline only ties with human supervision. AL provides no reliable advantage over random sampling under either LLM annotator. However, error structure varies sharply: only GPT-5.2 under the two-question interface produces classifiers with near-human FP/FN balance, while other LLM variants over-flag border-control and economic competition discourse. We release the dataset and code.

Large Language Models (LLMs) have advanced Table Question Answering, where most queries can be answered by extracting information or simple aggregation. However, a common class of real-world queries is implicitly predictive, requiring the inference of unobserved answers from historical patterns rather than mere retrieval. These queries introduce two challenges: recognizing latent intent and reliable predictive reasoning over massive tables. To assess LLMs in such Tabular questiOn answering with implicit Prediction tasks, we introduce TopBench, a benchmark consisting of 779 samples across four sub-tasks, ranging from single-point prediction to decision making, treatment effect analysis, and complex filtering, requiring models to generate outputs spanning reasoning text and structured tables. We evaluate diverse models under both text-based and agentic workflows. Experiments reveal that current models often struggle with intent recognition, defaulting to just lookups. Deeper analysis identifies that accurate intent disambiguation serves as the prerequisite for leading these predictive behaviors. Furthermore, elevating the upper bound of prediction precision requires the integration of more sophisticated modeling or reasoning capabilities.

AI agents such as OpenClaw are increasingly deployed in local workflows with access to external tools. This creates indirect prompt-injection (IPI) risk: an agent may execute harmful instructions embedded in untrusted inputs such as email, downloaded files, webpages, repositories, or group-chat messages. Existing evaluations are often small, purely simulated, or focused on a narrow set of channels. We introduce LivePI (Live Prompt Injection), a structured benchmark for IPI risk in a production-like but test-controlled environment. LivePI covers seven input surfaces, twelve attack/rendering families, and five malicious goals, including protected-information exfiltration, unauthorized security-control changes, unsafe code retrieval or execution, inbox-summary exfiltration, and cryptocurrency transfer. We run LivePI on a real virtual machine with live but test-controlled email, chat, web, local-file, repository, and wallet interfaces. Across GPT-5.3-Codex, Claude Opus 4.6, Gemini 3.1 Pro, Kimi K2.5, and GLM-5, total attack success rates range from 10.7% to 29.6%. Group-chat injection is uniformly successful across the evaluated backbones in our deployment, and repository-link attacks produce high-severity failures despite a small denominator. We also evaluate a two-layer defense consisting of prompt-level filtering and pre-execution tool-call authorization. In the GPT-5.3-Codex setting, the defense intercepts all tested malicious-goal completions in LivePI before execution while preserving benign utility on PinchBench-derived workloads.

AI coding agents are increasingly used for scientific work, but their end-to-end autonomous research capability remains difficult to verify. We present ResearchClawBench, a benchmark for evaluating autonomous scientific research across 40 tasks from 10 scientific domains. Each task is grounded in a real published paper, provides related literature and raw data, and hides the target paper during evaluation. Expert-curated multimodal rubrics decompose the target scientific artifacts into weighted criteria, enabling evaluation of target-paper-level re-discovery while leaving room for new discovery. We evaluate seven autonomous research (auto-research) agents under a unified protocol and seventeen native LLMs through the lightweight ResearchHarness. Current systems remain far from reliable re-discovery: the strongest autonomous agent, Claude Code, averages 21.5, and the strongest ResearchHarness LLM, Claude-Opus-4.7, averages 20.7, with an LLM frontier mean of only 26.5. Error analysis shows that failures concentrate in experimental protocol mismatch, evidence mismatch, and missing scientific core. ResearchClawBench provides a reproducible evaluation frontier for measuring progress toward autonomous scientific research.

With the rapid advancement of large language models (LLMs), aligning policy models with human preferences has become increasingly critical. Direct Preference Optimization (DPO) has emerged as a promising approach for alignment, acting as an RL-free alternative to Reinforcement Learning from Human Feedback (RLHF). Despite DPO's various advancements and inherent limitations, an in-depth review of these aspects is currently lacking in the literature. In this work, we present a comprehensive review of the challenges and opportunities in DPO, covering theoretical analyses, variants, relevant preference datasets, and applications. Specifically, we categorize recent studies on DPO based on key research questions to provide a thorough understanding of DPO's current landscape. Additionally, we propose several future research directions to offer insights on model alignment for the research community. An updated collection of relevant papers can be found on https://github.com/Mr-Loevan/DPO-Survey.

The estimation of origin-destination (OD) matrices is a crucial aspect of Intelligent Transport Systems (ITS). It involves adjusting an initial OD matrix by regressing the current observations like traffic counts of road sections (e.g., using least squares). However, the OD estimation problem lacks sufficient constraints and is mathematically underdetermined. To alleviate this problem, some researchers incorporate a prior OD matrix as a target in the regression to provide more structural constraints. However, this approach is highly dependent on the existing prior matrix, which may be outdated. Others add structural constraints through sensor data, such as vehicle trajectory and speed, which can reflect more current structural constraints in real-time. Our proposed method integrates deep learning and numerical optimization algorithms to infer matrix structure and guide numerical optimization. This approach combines the advantages of both deep learning and numerical optimization algorithms. The neural network(NN) learns to infer structural constraints from probe traffic flows, eliminating dependence on prior information and providing real-time performance. Additionally, due to the generalization capability of NN, this method is economical in engineering. We conducted tests to demonstrate the good generalization performance of our method on a large-scale synthetic dataset. Subsequently, we verified the stability of our method on real traffic data. Our experiments provided confirmation of the benefits of combining NN and numerical optimization.

We study city-scale control of electric-vehicle (EV) ride-hailing fleets where dispatch, repositioning, and charging decisions must respect charger and feeder limits under uncertain, spatially correlated demand and travel times. We formulate the problem as a hex-grid semi-Markov decision process (semi-MDP) with mixed actions -- discrete actions for serving, repositioning, and charging, together with continuous charging power -- and variable action durations. To guarantee physical feasibility during both training and deployment, the policy learns over high-level intentions produced by a masked, temperature-annealed actor. These intentions are projected at every decision step through a time-limited rolling mixed-integer linear program (MILP) that strictly enforces state-of-charge, port, and feeder constraints. To mitigate distributional shifts, we optimize a Soft Actor-Critic (SAC) agent against a Wasserstein-1 ambiguity set with a graph-aligned Mahalanobis ground metric that captures spatial correlations. The robust backup uses the Kantorovich-Rubinstein dual, a projected subgradient inner loop, and a primal-dual risk-budget update. Our architecture combines a two-layer Graph Convolutional Network (GCN) encoder, twin critics, and a value network that drives the adversary. Experiments on a large-scale EV fleet simulator built from NYC taxi data show that PD-RSAC achieves the highest net profit, reaching \$1.22M, compared with \$0.58M-\$0.70M for strong heuristic, single-agent RL, and multi-agent RL baselines, including Greedy, SAC, MAPPO, and MADDPG, while maintaining zero feeder-limit violations.

Time series (TS) reasoning models (TSRMs) have shown promising capabilities in general domains, yet they consistently fail in the financial domain, which exhibits unique characteristics. We propose a general 2 x 2 capability taxonomy for TSRMs by crossing 1) single-entity vs. multi-entity analysis with 2) assessment of the current state vs. prediction of future behavior. We instantiate this taxonomy in the financial domain-where the distinction between deterministic assessment and stochastic prediction is particularly critical-as ten financial reasoning tasks, forming the FinTSR-Bench benchmark based on S&P stocks. To this end, we propose FinSTaR (Financial Time Series Thinking and Reasoning), trained on FinTSR-Bench with distinct chain-of-thought (CoT) strategies tailored to each category. For assessment, which is deterministic (i.e., computable from observable data), we employ Compute-in-CoT, a programmatic CoT that enables models to derive answers directly from raw prices. For prediction, which is inherently stochastic (i.e., subject to unobservable factors), we adopt Scenario-Aware CoT, which generates diverse scenarios before making a judgment, mirroring how financial analysts reason under uncertainty. The proposed method achieves 78.9% average accuracy on FinTSR-Bench, substantially outperforming LLM and TSRM baselines. Furthermore, we show that the four capability categories are complementary and mutually reinforcing through joint training, and that Scenario-Aware CoT consistently improves prediction accuracy over standard CoT. Code is available at https://github.com/seunghan96/FinSTaR.

A scientific hypothesis is the first step in research and undergoes experimental validation, yet it also reflects a deep understanding of and reasoning about scientific phenomena. We introduce DN-Hypo-Pipeline, an AI-powered workflow based on large language models, designed to support structured scientific thinking and hypothesis generation by leveraging scientific explanations as prior knowledge. This pipeline assists researchers in deriving novel hypotheses from existing literature. Given the explanandum (i.e., the conclusion) of a research paper, it identifies underlying laws, theories, and principles, and reconstructs a new, yet-to-be-verified explanation for the observed phenomenon. We evaluated DN-Hypo-Pipeline in the field of data science modeling using three highly cited papers. Statistical inference, supported by both LLM-as-judge assessment and human expert evaluation, demonstrates that our pipeline is more effective than direct generation methods. Additionally, we validated the two highest-scoring generated hypotheses by developing corresponding novel algorithms, which outperformed the baseline models presented in the original papers. Beyond application in data science, DN-Hypo-Pipeline provides a theoretical framework that not only encompasses theory-guided data science modeling methods but also reveals a more fundamental structure of the modeling process. Moreover, this approach is essentially a generalization of theory-guided modeling, offering potential for extension to other domains and across a broader range of scientific disciplines.

Cancer treatment is at the core a sequential decision-making problem with partial observability, latent patient heterogeneity, and explicit constraints on the budget for medical measurements. Unlike standard Reinforcement Learning (RL) approaches that control state trajectories, cancer treatments permanently modify patients' transition dynamics, changing how states evolve over time. We model cancer treatment as a belief-space planning problem using active inference, deriving an expected free-energy objective that unifies goal-directed control and information acquisition under measurement budgets without. We implement this framework using real clinical cancer data from the AACR Project GENIE Biopharma Collaborative dataset. Results on clinical data demonstrate a simultaneous patient categorization and high treatment efficacy, under real measurement and treatment constraints.

Estimating queue lengths at signalized intersections is a long-standing challenge in traffic management. Partial observability of vehicle flows complicates this task despite the availability of two privacy-preserving data sources: (i) aggregated vehicle counts from loop detectors near stop lines, and (ii) aggregated floating car data (aFCD) that provide segment-wise average speed measurements. However, how to integrate these sources with differing spatial and temporal resolutions for queue length estimation is rather unclear. Addressing this question, we present Q-Net: a queue estimation framework built upon a state-space formulation. This design addresses key challenges in queue modeling, such as violations of traffic conservation assumptions. Q-Net follows the Kalman predict-update structure and maintains physical interpretability in both the state evolution and measurement models. Q-Net uses an AI-augmented Kalman filter to learn time-varying gain dynamics from data. The framework supports real-time implementation and improves spatial transferability by grouping aFCD measurements into fixed-size local groups, making the number of learnable parameters independent of section length. Evaluations on urban main roads in Rotterdam, the Netherlands, show that Q-Net outperforms baseline methods, tracks queue formation and dissipation accurately, and mitigates aFCD-induced delays. By combining data efficiency, interpretability, real-time applicability, and spatial transferability, Q-Net makes accurate queue length estimation possible without costly sensing infrastructure like cameras or radar.

OD matrix estimation is a critical problem in the transportation domain. The principle method uses the traffic sensor measured information such as traffic counts to estimate the traffic demand represented by the OD matrix. The problem is divided into two categories: static OD matrix estimation and dynamic OD matrices sequence(OD sequence for short) estimation. The above two face the underdetermination problem caused by abundant estimated parameters and insufficient constraint information. In addition, OD sequence estimation also faces the lag challenge: due to different traffic conditions such as congestion, identical vehicle will appear on different road sections during the same observation period, resulting in identical OD demands correspond to different trips. To this end, this paper proposes an integrated method, which uses deep learning methods to infer the structure of OD sequence and uses structural constraints to guide traditional numerical optimization. Our experiments show that the neural network(NN) can effectively infer the structure of the OD sequence and provide practical constraints for numerical optimization to obtain better results. Moreover, the experiments show that provided structural information contains not only constraints on the spatial structure of OD matrices but also provides constraints on the temporal structure of OD sequence, which solve the effect of the lagging problem well.

Fatigue detection for human operators is important in safety-related applications such as aviation, mining, and long-haul transport. Reliable estimation of operator fatigue can support timely warnings, adaptive task scheduling, takeover reminders, and other safety-management decisions in human-machine systems. However, the effectiveness of these functions depends on whether fatigue-related signals can be reliably captured in the deployment environment. While many studies have shown the value of high-fidelity sensors in controlled laboratory environments, their performance often degrades when used in real-world settings because of noise, lighting conditions, and field-of-view constraints, thereby limiting their practical use. This paper formalizes a deployment-oriented setting for real-world fatigue detection, where high-quality sensors are often unavailable in practical applications. To address this issue, we use knowledge from heterogeneous source domains, including high-fidelity sensors that are difficult to deploy in the field but commonly used in controlled environments, to assist fatigue detection in the real-world target domain. Based on this idea, we design a heterogeneous and multi-source fatigue-detection framework that uses the available modalities in the target domain while leveraging diverse configurations in the source domains through cross-domain modality imputation based on shared modalities.

Decoded Neurofeedback (DecNef) is a promising non-invasive approach to brain modulation with wide-ranging applications in neuromedicine and cognitive neuroscience. However, progress in DecNef research remains constrained by subject-dependent learning variability, reliance on indirect measures to quantify progress, and the high cost and time demands of experimentation. We present DecNefSimulator, a modular and interpretable simulation framework that formalizes DecNef as a machine learning problem. Beyond providing a virtual laboratory, DecNefSimulator enables researchers to model, analyze and understand neurofeedback dynamics. Using latent variable generative models as simulated participants, DecNefSimulator allows direct observation of internal cognitive states and systematic evaluation of how different protocol designs and subject characteristics influence learning. We demonstrate how this approach can (i) reproduce empirical phenomena of DecNef learning, (ii) identify conditions under which DecNef feedback fails to induce learning, and (iii) guide the design of more robust and reliable DecNef protocols in silico before human implementation. In summary, DecNefSimulator bridges computational modeling and cognitive neuroscience, offering a principled foundation for methodological innovation, robust protocol design, and ultimately, a deeper understanding of DecNef-based brain modulation.

Understanding disease progression is a central clinical challenge with direct implications for early diagnosis and personalized treatment. While recent generative approaches have attempted to model progression, key mismatches remain: disease dynamics are inherently continuous and monotonic, yet latent representations are often scattered, lacking semantic structure, and diffusion-based models disrupt continuity with random denoising process. In this work, we propose to treat the disease dynamic as a velocity field and leverage Flow Matching (FM) to align the temporal evolution of patient data. Unlike prior methods, it captures the intrinsic dynamic of disease, making the progression more interpretable. However, a key challenge remains: in latent space, Auto-Encoders (AEs) do not guarantee alignment across patients or correlation with clinical-severity indicators (e.g., age and disease conditions). To address this, we propose to learn patient-specific latent alignment, which enforces patient trajectories to lie along a specific axis, with magnitude increasing monotonically with disease severity. This leads to a consistent and semantically meaningful latent space. Together, we present $Δ$-LFM, a framework for modeling patient-specific latent progression with flow matching. Across three longitudinal MRI benchmarks, $Δ$-LFM demonstrates strong empirical performance and, more importantly, offers a new framework for interpreting and visualizing disease dynamics.

We present DeepInflation, an AI agent designed for research and model discovery in inflationary cosmology. Built upon a multi-agent architecture, DeepInflation integrates Large Language Models (LLMs) with a symbolic regression (SR) engine and a retrieval-augmented generation (RAG) knowledge base. This framework enables the agent to automatically explore and verify the vast landscape of inflationary potentials while grounding its outputs in established theoretical literature. We demonstrate that DeepInflation can successfully discover simple and viable single-field slow-roll inflationary potentials consistent with the latest observations (with the ACT DR6 results taken as an example) or any given $n_s$ and $r$, and provide accurate theoretical context for obscure inflationary scenarios. DeepInflation serves as a prototype for a new generation of autonomous scientific discovery engines in cosmology, which enables researchers and non-experts alike to explore the inflationary landscape using natural language. This agent is available at https://github.com/pengzy-cosmo/DeepInflation.

Small and Medium Enterprises (SMEs) constitute 99.9% of U.S. businesses and generate 44% of economic activity, yet systematically identifying high-potential SMEs remains an open challenge. We introduce SME-HGT, a Heterogeneous Graph Transformer framework that predicts which SBIR Phase I awardees will advance to Phase II funding using exclusively public data. We construct a heterogeneous graph with 32,268 company nodes, 124 research topic nodes, and 13 government agency nodes connected by approximately 99,000 edges across three semantic relation types. SME-HGT achieves an AUPRC of 0.621 0.003 on a temporally-split test set, outperforming an MLP baseline (0.590 0.002) and R-GCN (0.608 0.013) across five random seeds. At a screening depth of 100 companies, SME-HGT attains 89.6% precision with a 2.14 lift over random selection. Our temporal evaluation protocol prevents information leakage, and our reliance on public data ensures reproducibility. These results demonstrate that relational structure among firms, research topics, and funding agencies provides meaningful signal for SME potential assessment, with implications for policymakers and early-stage investors.

We present a physics-based neural network framework for the discovery of constitutive models in fully coupled thermomechanics. In contrast to classical formulations based on the Helmholtz energy, we adopt the internal energy and a dissipation potential as primary constitutive functions, expressed in terms of deformation and entropy. This choice avoids the need to enforce mixed convexity--concavity conditions and facilitates a consistent incorporation of thermodynamic principles. In this contribution, we focus on materials without preferred directions or internal variables. While the formulation is posed in terms of entropy, the temperature is treated as the independent observable, and the entropy is inferred internally through the constitutive relation, enabling thermodynamically consistent modeling without requiring entropy data. Thermodynamic admissibility of the networks is guaranteed by construction. The internal energy and dissipation potential are represented by input convex neural networks, ensuring convexity and compliance with the second law. Objectivity, material symmetry, and normalization are embedded directly into the architecture through invariant-based representations and zero-anchored formulations. We demonstrate the performance of the proposed framework on synthetic and experimental datasets, including purely thermal problems and fully coupled thermomechanical responses of soft tissues and filled rubbers. The results show that the learned models accurately capture the underlying constitutive behavior. All code, data, and trained models are made publicly available via https://doi.org/10.5281/zenodo.19248596.

Context: Schools, training platforms, and technology firms increasingly need to assess programming proficiency at scale with transparent, reproducible methods that support personalized learning pathways. Objective: This study introduces a pedagogical framework for Scratch project assessment, aligned with the Common European Framework of Reference (CEFR), providing universal competency levels for students and teachers alongside actionable insights for curriculum design. Method: We apply Fuzzy C-Means clustering to 2008246 Scratch projects evaluated via Dr.Scratch, implementing an ordinal criterion to map clusters to CEFR levels (A1-C2), and introducing enhanced classification metrics that identify transitional learners, enable continuous progress tracking, and quantify classification certainty to balance automated feedback with instructor review. Impact: The framework enables diagnosis of systemic curriculum gaps-notably a "B2 bottleneck" where only 13.3% of learners reside due to the cognitive load of integrating Logic Synchronization, and Data Representation--while providing certainty--based triggers for human intervention.

Effective adaptation and mitigation strategies for climate change require high-resolution projections to inform strategic decision-making. Conventional global climate models, which typically operate at resolutions of 150 to 200 kilometers, lack the capacity to represent essential regional processes. IPSL-AID is a global to regional downscaling tool based on a denoising diffusion probabilistic model designed to address this limitation. Trained on ERA5 reanalysis data, it generates 0.25 degree resolution fields for temperature, wind, and precipitation using coarse inputs and their spatiotemporal context. It also models probability distributions of fine-scale features to produce plausible scenarios for uncertainty quantification. The model accurately reconstructs statistical distributions, including extreme events, power spectra, and spatial structures. This work highlights the potential of generative diffusion models for efficient climate downscaling with uncertainty

Large language models can write scientific code, but direct paper-to-program translation remains fragile when correctness depends on tacit conventions rather than explicit equations. We frame this as a \textbf{knowledge-externalization} problem: index choices, gauges, fermionic signs, contraction order, validation gates, and scaling constraints must be made explicit before code generation. We evaluate a multi-stage, human-in-the-loop workflow on two quantum many-body tasks. DMRG from Schollwock's pedagogical review serves as calibration: specification-guided implementations pass in all 16 model pairings, compared with 6/13 direct attempts, and a prose-specification ablation shows that externalized content, not \LaTeX{} form, is the active ingredient. Pfaffian conversion of HFB states to MPS from the five-page Letter by Jin et al. serves as the stress test: no public implementation is available, and success depends on tacit sign, gauge, ordering, and scalability conventions. Here the workflow yields 11/26 audited passes, while direct prompting yields none. Cross-specification transfer is asymmetric: non-GPT specifications implemented by GPT~5.5 pass 4/4, whereas GPT~5.5 specifications implemented by weaker models fail 4/4. The contrast supports a two-bottleneck picture. Externalization resolves the first bottleneck -- paper-to-code ambiguity -- well enough to make DMRG reproducible and Pfaffian-MPS auditable. The remaining failures expose a second bottleneck in implementation-model capability. Iterative meta-specification moves this boundary but does not eliminate it. The resulting \emph{Paper-to-Program Many-Body} skill is both a reusable implementation protocol and a diagnostic instrument for AI-assisted many-body programming.

The inherent electronic and speckle noise complicates clinical interpretation of ultrasound images. Conventional denoising methods rely on explicit noise assumptions whose validity diminishes under composite noise conditions. Learning-based methods are usually pretrained in a limited image domain using a labeled dataset, which implies inevitable domain shift in complex in vivo environments. This study proposes a Pyramid Self-Contrastive Learning (PSCL) framework for test-time ultrasound image denoising without pretraining. Given multiple noisy samples from only one-shot imaging, PSCL disentangles anatomical similarity and noise randomness into separate pyramid latent spaces. The clean image is then decoded from the anatomy space while discarding the noise space. We first apply PSCL to synthetic aperture ultrasound (SAU), where an Aperture-to-Aperture loop serves as a self-supervised proxy task to ensure denoising fidelity. Simulation experiments, including noise levels from 0 to 30 dB and inclusion geometries from simple to complex, demonstrated improvements of 69.3% in SNR and 34.4% in CNR. The in vivo results showed 84.8% SNR and 25.7% CNR gains using only two aperture data of the heart in six echocardiographic views, liver, and kidney. PSCL delivers clear images across diverse imaging targets and configurations, paving the way for more reliable anatomical visualization without domain shift and pretraining costs.

Accurate disease risk prediction is challenged by heterogeneous features, limited data, and class imbalance. This study presents yvsoucom-iterkit, a deterministic AutoML framework that models pipeline optimization as a configuration-level system with full reproducibility and traceable execution logs, enabling systematic analysis of component attribution, interactions, similarity, and cross-seed robustness. Experiments on the Pima Indians Diabetes and Stroke datasets across more than 18,000 pipeline configurations reveal a structured yet partially redundant search space, where performance is dominated by a small subset of interacting components. Ensemble models achieve stable performance, reaching a Weighted-F1 of 0.89 on Pima and 0.94 on Stroke. Macro-F1 reaches approximately 0.88 on Pima but drops to 0.6560 on Stroke due to severe imbalance. Cross-seed experiments show that ensembles reduce variance compared to single models. Friedman testing ($p < 0.05$) confirms significant ranking differences across configurations. Based on analysis of component attribution, interaction, and similarity, optimal configuration design reveals dataset-dependent behavior. For the Pima dataset, computational efficiency benefits from simplified search spaces where redundant components can be removed, with split ratio playing a key role. In contrast, the Stroke dataset requires enhanced imbalance-aware strategies, where RandomOverSampler improves Macro-F1 from 0.6560 to 0.6766. These findings demonstrate that effective AutoML optimization is achieved through optimal configuration design, where carefully constraining the search space to high-impact components can improve performance, stability, and interpretability while reducing unnecessary search complexity.

Deep learning-based CT segmentation systems often achieve high accuracy on clean benchmark images, but their performance may degrade under heterogeneous clinical imaging conditions such as noise, resolution loss, contrast variation, intensity shift, and artifacts. This instability can limit reliable deployment in real-world medical imaging workflows. We propose Robustness via Augmented Multi-corruption Pipeline (RAMP), a robustness-oriented augmentation framework for CT segmentation. RAMP combines anatomically constrained spatial perturbations, CT intensity transformations, and stochastic multi-corruption composition to expose models to clinically plausible image degradation during training. Across two CT segmentation evaluation settings, RAMP achieved the strongest corrupted-image performance and the smallest clean-to-corrupted robustness gap. In the five-organ noisy evaluation benchmark, RAMP improved mean corrupted Dice from 0.610 to 0.753 and reduced the robustness gap from 0.264 to 0.064 compared with the nnU-Net baseline. In Abdomen1K, RAMP improved mean corrupted Dice from 0.633 to 0.789 and reduced the robustness gap from 0.290 to 0.070. Although RAMP did not achieve the highest clean-image Dice, it substantially mitigated worst-case segmentation collapse under severe image degradation. These results suggest that multi-corruption augmentation can serve as a practical pre-deployment strategy for improving the reliability of CT segmentation systems in heterogeneous clinical environments.

Artificial intelligence provides a practical framework for crop damage assessment from imagery data, supporting early decision-making in agricultural management. In peach orchards, climate change increases abiotic stress and biotic pressures, including pests and diseases, which often produce visually similar foliar symptoms. This overlap makes manual diagnosis difficult, especially across multiple fields with varying environmental conditions, highlighting the need for automated models with strong generalization ability. We propose an image-based classification approach for peach leaf damage detection. A benchmark dataset was created through manual annotation of publicly available images, consisting of 1,366 peach leaves across six damage categories. Several deep learning architectures were evaluated. EfficientNet models achieved the best results, with EfficientNetB0 reaching 92.9 percent accuracy, EfficientNetB3 achieving 91.5 percent, and EfficientNetB5 showing the strongest performance on minority classes. DenseNet121 reached 92.6 percent accuracy. The integration of the Convolutional Block Attention Module (CBAM) improved performance in several backbones, particularly EfficientNetB5 and InceptionV3, while showing limited or negative impact in others. The CBAM-enhanced EfficientNetB5 achieved the best overall accuracy of 93.3 percent. To evaluate robustness under realistic conditions, a local dataset of 180 images across four classes was collected, and transfer learning strategies were applied to address domain shift. Three fine-tuning strategies were tested. EfficientNetB3 combined with CBAM achieved the best performance in the local domain, reaching a 93 percent macro F1-score after transfer. Overall, attention-based models showed improved robustness for minority classes and better generalization across different field conditions.

In-silico trials of medical devices require the generation of virtual populations of anatomies. In cardiovascular applications, virtual anatomy is typically represented as a 3D+t mesh sampled from a generative model. However, most existing mesh generators focus on static anatomy, while sequence models often lack explicit periodicity. To this end, we propose 4D F-MeshLDM, a conditional generative framework comprising a convolutional mesh VAE to encode meshes, a structural latent space that parameterises motion using a truncated Fourier series, and a diffusion prior that learns the latent distribution over Fourier coefficient tokens. By conditioning the diffusion process on clinical covariates via affine modulation, we enable controllable synthesis. Sampling tokens and performing inverse Fourier synthesis yield cycle-consistent latent trajectories, which can be decoded into 3D+t cardiac mesh sequences. Experiments on 5,000 UK Biobank subjects demonstrate that 4D F-MeshLDM outperforms state-of-the-art baselines in anatomical fidelity and achieves near-zero cycle closure error. Furthermore, the generated cohorts accurately preserve clinical functional indices, highlighting the potential of our framework for reliable in-silico cardiac trials.

Information-theoretic (IT) measures are ubiquitous in artificial intelligence: entropy drives decision-tree splits and uncertainty quantification, cross-entropy is the default classification loss, mutual information underpins representation learning and feature selection, and transfer entropy reveals directed influence in dynamical systems. A second, less consolidated family of measures, integrated information (Phi), effective information (EI), and autonomy, has emerged for characterizing agent complexity. Despite wide adoption, measure selection is often decoupled from estimator assumptions, failure modes, and safe inferential claims. This paper provides a practical decision framework for all seven measures, organized around three prescriptive questions for each: (i) what question does the measure answer and in which AI context; (ii) which estimator is appropriate for the data type and dimensionality; and (iii) what is the most dangerous misuse. The framework is operationalized in two complementary artifacts: a measure-selection flowchart and a master decision table. We cover both AI/ML and decision-making agent application domains per measure, with standardized Bridge Boxes linking IT quantities to cognitive constructs. Three worked examples illustrate the framework on concrete practitioner scenarios spanning representation learning, temporal influence analysis, and evolved agent complexity.

Artificial intelligence in France is often discussed through separate dimensions such as investment, compute, regulation, employment, sovereignty, and education. This viewpoint paper proposes a unified interpretation: France can be analyzed as a national AI learning system. Building on Human-Centered Learning Mechanics (HCLM), we use HCLM not as a validated econometric model, but as a conceptual and diagnostic lens for interpreting national AI development as a balance between information injection, absorptive capacity, and institutional dissipation. Information injection includes compute, data, talent, research, capital, industrial deployment, and policy experimentation. Institutional dissipation refers to avoidable frictions such as administrative overload, coordination failures, energy constraints, regulatory uncertainty, talent mobility pressures, and weak industrial absorption. Regulation is not treated as mere friction: adaptive governance, trusted data spaces, and safety-oriented standards may increase long-term learning capacity by improving legitimacy, interoperability, and social trust. The central claim is not that a country follows neural-network equations, but that AI sovereignty depends on how effectively it converts distributed information into absorbed, coordinated, and socially legitimate capability. The paper connects HCLM with neural scaling laws, endogenous growth theory, creative destruction, absorptive capacity, and coordination mechanisms. It offers a formal heuristic, policy indicators, illustrative scenarios, and implications for France. The numerical results are diagnostic scenarios, not econometric estimates or official rankings. The proposed viewpoint reframes AI policy as the governance of an open, strategic, non-equilibrium learning system that should be tested with historical and cross-country data.

Point cloud stands as the most widely adopted format for representing 3D shapes and scenes due to its simplicity and geometric fidelity. However, its inherent unordered and irregular nature, exacerbated by sensor noise and occlusions, introduces unique challenges for machine learning based methodologies. To combat these issues, diverse strategies have been developed, including converting to a format that has orderliness, extracting local geometry, and permutation-invariant or self-attention-based processing. In this paper, our focus is directed towards deep learning models for three fundamental tasks in 3D vision: point cloud classification, part segmentation, and semantic segmentation. We begin by formally defining point cloud data, followed by an in-depth discussion on its structural characteristics. Then, we categorize notable works based on their backbone structure and evaluate their performance on popular benchmarks. Beyond empirical comparison, we offer insights into architectural innovations and limitations. We also outline open challenges and promising future directions for 3D point cloud understanding.

Human-AI collaboration is considered the most promising way to incorporate AI in the workplace. What remains unexplored are the experiential consequences of this teaming. More specifically, in a team with AI, how humans perceive themselves (self-perception) and how they are perceived by their coworkers (peer perception) in terms of work ownership and job meaningfulness. In a 2x2x2 vignette study (n=50), participants rated perceptions of ownership, affect, job meaningfulness and satisfaction, and role dynamics across two levels (low/high) of AI proactivity and AI competency as within-subject factors, with point-of-view (self perception/peer perception) as between-subjects. Our results showed that AI with low competency or low proactivity generally improved feelings related to ownership, meaningfulness, satisfaction, and role dynamics, and also increased positive affect while reducing negative affect. However, these effects were often influenced by point-of-view. For instance, low AI proactivity resulted in higher job satisfaction from self-perception rather than peer perception. Based on our findings, we argue that designing AI for the future of work solely around performance metrics may not be adequate. Highly competent and proactive AI-driven systems can have undesirable impacts on perceptions of ownership, job identity, social image and team dynamics, and consequently, job meaningfulness.

Millions of individuals worldwide suffer from sensory and communication deficits caused by neurodegenerative diseases, stroke, or trauma. Brain-computer interfaces (BCIs) offer a promising avenue for sensory and motor restoration. However, the scientific literature remains highly fragmented between invasive neuroprosthetics and non-invasive electrophysiological decoders, with a lack of consistent terminology and comparison metrics. This chapter proposes a unified 2 x 2 framework categorizing BCIs along two axes: degree of invasiveness (invasive vs. non-invasive) and signal direction (afferent sensory-IN vs. efferent sensory-OUT). We define and distinguish the paradigms of restoration, substitution, and augmentation. Furthermore, we outline a structural roadmap for the convergence of these modalities over near-, medium-, and long-term horizons, focusing on physical limits and the integrative role of machine learning foundation models.

Deploying Multimodal Large Language Models as the brain of embodied agents remains challenging, particularly under long-horizon observations and limited context budgets. Existing memory assisted methods often rely on textual summaries, which discard rich visual and spatial details and remain brittle in non-stationary environments. In this work, we propose a non-parametric memory framework that explicitly disentangles episodic and semantic memory for embodied exploration and question answering. Our retrieval-first, reasoning-assisted paradigm recalls episodic experiences via semantic similarity and verifies them through visual reasoning, enabling robust reuse of past observations without rigid geometric alignment. In parallel, we introduce a program-style rule extraction mechanism that converts experiences into structured, reusable semantic memory, facilitating cross-environment generalization. Extensive experiments demonstrate state-of-the-art performance on embodied question answering and exploration benchmarks, yielding a 7.3% gain in LLM-Match and an 11.4% gain in LLM MatchXSPL on A-EQA, as well as +7.7% success rate and +6.8% SPL on GOAT-Bench. Analyses reveal that our episodic memory primarily improves exploration efficiency, while semantic memory strengthens complex reasoning of embodied agents.

With the rise of large language models (LLMs), they have become instrumental in applications such as Retrieval-Augmented Generation (RAG). Yet evaluating these systems remains bottlenecked by the time and cost of building specialized assessment datasets. We introduce KNIGHT, an LLM-based, knowledge-graph-driven framework for generating multiple-choice question (MCQ) datasets from external sources. KNIGHT constructs a topic-specific knowledge graph, a structured and parsimonious summary of entities and relations, that can be reused to generate instructor-controlled difficulty levels, including multi-hop questions, without repeatedly re-feeding the full source text. This knowledge graph acts as a compressed, reusable state, making question generation a cheap read over the graph. We instantiate KNIGHT on Wikipedia/Wikidata while keeping the framework domain- and ontology-agnostic. As a case study, KNIGHT produces six MCQ datasets in History, Biology, and Mathematics. We evaluate quality on five criteria: fluency, unambiguity (single correct answer), topic relevance, option uniqueness, and answerability given the provided sources (as a proxy for hallucination). Results show that KNIGHT enables token- and cost-efficient generation from a reusable graph representation, achieves high quality across these criteria, and yields model rankings aligned with MMLU-style benchmarks, while supporting topic-specific and difficulty-controlled evaluation.

A data-driven inverse optimization problem (DDIOP) is the problem of estimating the objective-function parameters (weights) that explain observed optimal-solution data, and it arises in many applications, including mixed integer linear programming (MILP). In inverse optimization for MILPs, the prediction error of the features is discontinuous with respect to the weights, so applying gradient-based optimization directly is difficult. In this paper we focus on the suboptimality loss. This loss attains its minimum value, zero, if and only if the weights are exactly consistent with the observed data. We reveal a geometric structure of this loss -- it is convex and piecewise linear, and moreover the set of weights that are exactly consistent with the observed data has a positive ``thickness'' rather than being a single point or a thin boundary -- and use it to show the following. First, a broad class of gradient-based optimization methods, including projected subgradient descent, reaches exact consistency with the observed data in finitely many iterations (an exact solution is obtained in finite time). Second, for projected subgradient descent we give an explicit upper bound on the number of iterations needed to reach exact consistency. Third, when the forward problem is an integer linear program (ILP), we give this upper bound as a fully explicit iteration count determined solely by the number of samples, the dimension of the features, and the structure of the constraint coefficient matrix. Through numerical experiments, we confirm this finite-step attainment behavior.

A fully tensorial theoretical framework for hypercomplex-valued neural networks is presented. The proposed approach enables neural network architectures to operate on data defined over arbitrary finite-dimensional algebras. The central observation is that algebra multiplication can be represented by a rank-three tensor, which allows all algebraic operations in neural network layers to be formulated in terms of standard tensor contractions, permutations, and reshaping operations. This tensor-based formulation provides a unified and dimension-independent description of hypercomplex-valued dense and convolutional layers and is directly compatible with modern deep learning libraries supporting optimized tensor operations. The proposed framework recovers existing constructions for four-dimensional algebras as a special case. Within this setting, a tensor-based version of the universal approximation theorem for single-layer hypercomplex-valued perceptrons is established under mild non-degeneracy assumptions on the underlying algebra, thereby providing a rigorous theoretical foundation for the considered class of neural networks.

As artificial intelligence (AI) becomes increasingly integrated into education, understanding how students perceive its risks is essential for supporting responsible and effective adoption. This research aimed to examine the relationships between perceived AI competence and risks among Finnish K-12 upper secondary students (n = 163) by utilizing a co-occurrence analysis. Students reported their self-perceived AI competence and concerns related to AI across systemic, institutional, and personal domains. The findings showed that students with lower competence emphasized personal and learning-related risks, such as reduced creativity, lack of critical thinking, and misuse, whereas higher-competence students focused more on systemic and institutional risks, including bias, inaccuracy, and cheating. These differences suggest that students' self-reported AI competence is related to how they evaluate both the risks and opportunities associated with artificial intelligence in education (AIED). The results of this study highlight the need for educational institutions to incorporate AI literacy into their curricula, provide teacher guidance, and inform policy development to ensure personalized opportunities for utilization and equitable integration of AI into K-12 education.

Retrieval-Augmented Generation (RAG) systems are emerging as a key approach for grounding Large Language Models (LLMs) in external knowledge, addressing limitations in factual accuracy and contextual relevance. However, there is a lack of empirical studies that report on the development of RAG-based implementations grounded in real-world use cases, evaluated through general user involvement, and accompanied by systematic documentation of lessons learned. This paper presents five domain-specific RAG applications developed for real-world scenarios across governance, cybersecurity, agriculture, industrial research, and medical diagnostics. Each system incorporates multilingual OCR, semantic retrieval via vector embeddings, and domain-adapted LLMs, deployed through local servers or cloud APIs to meet distinct user needs. A web-based evaluation involving a total of 100 participants assessed the systems across six dimensions: (i) Ease of Use, (ii) Relevance, (iii) Transparency, (iv) Responsiveness, (v) Accuracy, and (vi) Likelihood of Recommendation. Based on user feedback and our development experience, we documented twelve key lessons learned, highlighting technical, operational, and ethical challenges affecting the reliability and usability of RAG systems in practice.

Prompt optimization aims to search for effective prompts that enhance the performance of large language models (LLMs). Although existing prompt optimization methods have discovered effective prompts, they often differ from sophisticated prompts carefully designed by human experts. Prompt design strategies, representing best practices for improving prompt performance, can be key to improving prompt optimization. Recently, a method termed the Autonomous Prompt Engineering Toolbox (APET) has incorporated various prompt design strategies into the prompt optimization process. In APET, the LLM is needed to implicitly select and apply the appropriate strategies because prompt design strategies can have negative effects. This implicit selection may be suboptimal due to the limited optimization capabilities of LLMs. This paper introduces Optimizing Prompts with sTrategy Selection (OPTS), which implements explicit selection mechanisms for prompt design. We propose three mechanisms, including a Thompson sampling-based approach, and integrate them into EvoPrompt, a well-known prompt optimizer. Experiments optimizing prompts for two LLMs, Llama-3-8B-Instruct and GPT-4o mini, were conducted using BIG-Bench Hard. Our results show that the selection of prompt design strategies improves the performance of EvoPrompt, and the Thompson sampling-based mechanism achieves the best overall results. Our experimental code is provided at https://github.com/shiralab/OPTS .

In this short note, we report and analyze a striking event: OpenAI's large language model o3 has outwitted all students in a university exam on thermodynamics. The thermodynamics exam is a difficult hurdle for most students, where they must show that they have mastered the fundamentals of this important topic. Consequently, the failure rates are very high, A-grades are rare - and they are considered proof of the students' exceptional intellectual abilities. This is because pattern learning does not help in the exam. The problems can only be solved by knowledgeably and creatively combining principles of thermodynamics. We have given our latest thermodynamics exam not only to the students but also to OpenAI's most powerful reasoning model, o3, and have assessed the answers of o3 exactly the same way as those of the students. In zero-shot mode, the model o3 solved all problems correctly, better than all students who took the exam; its overall score was in the range of the best scores we have seen in more than 10,000 similar exams since 1985. This is a turning point: machines now excel in complex tasks, usually taken as proof of human intellectual capabilities. We discuss the consequences this has for the work of engineers and the education of future engineers.

Large language models (LLMs) have shown increasing promise in educational settings, yet their mathematical reasoning has been considered evolving. This study evaluates the mathematical capabilities of various LLMs using the Finnish matriculation examination, a high-stakes digital test for upper secondary education. Initial tests yielded moderate performance corresponding to mid-range grades, but later evaluations demonstrated substantial improvements as the language models evolved. Remarkably, some models achieved near-perfect or perfect scores, matching top student performance and qualifying for university admission. Our findings highlight the rapid advances in the mathematical proficiency of LLMs and illustrate their potential as underlying tools to support learning and teaching in a variety of ways.

Cyber-Physical Systems (CPS) are abundant in safety-critical domains such as healthcare, avionics, and autonomous vehicles. Formal verification of their operational safety is, therefore, of utmost importance. In this paper, we address the falsification problem, where the focus is on searching for an unsafe execution in the system instead of proving their absence. The contribution of this paper is a framework that (a) connects the falsification of CPS with the falsification of deep neural networks (DNNs) and (b) leverages the inherent interpretability of Decision Trees for faster falsification of CPS. This is achieved by: (1) building a surrogate model of the CPS under test, either as a DNN model or a Decision Tree, (2) application of various DNN falsification tools to falsify CPS, and (3) a novel falsification algorithm guided by the explanations of safety violations of the CPS model extracted from its Decision Tree surrogate. The proposed framework has the potential to exploit a repertoire of \emph{adversarial attack} algorithms designed to falsify robustness properties of DNNs, as well as state-of-the-art falsification algorithms for DNNs. Although the presented methodology is applicable to systems that can be executed/simulated in general, we demonstrate its effectiveness, particularly in CPS. We show that our framework, implemented as a tool \textsc{FlexiFal}, can detect hard-to-find counterexamples in CPS that have linear and non-linear dynamics. Decision tree-guided falsification shows promising results in efficiently finding multiple counterexamples in the ARCH-COMP 2024 falsification benchmarks~\cite{khandait2024arch}.

Music is a universal phenomenon that profoundly influences human experiences across cultures. This study investigates whether music can be decoded from human brain activity measured with functional MRI (fMRI) during its perception. Leveraging recent advancements in extensive datasets and pre-trained computational models, we construct mappings between neural data and latent representations of musical stimuli. Our approach integrates functional and anatomical alignment techniques to facilitate cross-subject decoding, addressing the challenges posed by the low temporal resolution and signal-to-noise ratio (SNR) in fMRI data. Starting from the GTZan fMRI dataset, where five participants listened to 540 musical stimuli from 10 different genres while their brain activity was recorded, we used the CLAP (Contrastive Language-Audio Pretraining) model to extract latent representations of the musical stimuli and developed voxel-wise encoding models to identify brain regions responsive to these stimuli. By applying a threshold to the association between predicted and actual brain activity, we identified specific regions of interest (ROIs) which can be interpreted as key players in music processing. Our decoding pipeline, primarily retrieval-based, employs a linear map to project brain activity to the corresponding CLAP features. This enables us to predict and retrieve the musical stimuli most similar to those that originated the fMRI data. Our results demonstrate state-of-the-art identification accuracy, with our methods significantly outperforming existing approaches. Our findings suggest that neural-based music retrieval systems could enable personalized recommendations and therapeutic applications. Future work could use higher temporal resolution neuroimaging and generative models to improve decoding accuracy and explore the neural underpinnings of music perception and emotion.