As large language models increasingly deployed into agentic systems, existing methods face critical gaps in observing, assessing, and mitigating deployment-specific risks. We present a comprehensive, observability-driven workflow: we introduce \textbf{AgentSeer}, observability tool which decomposes agentic executions into granular \emph{action-component} graphs; we use this decomposition to rigorously quantify the gap between model-level and agent-level jailbreaking risk via cross-model validation on GPT-OSS-20B and Gemini-2.0-flash with HarmBench under single-turn and iterative-refinement attacks; we leverage action-graph risk signals to automate iterative prompt hardening against direct and iterative jailbreak attacks. Stark differences is revealed between model-level and agentic-level vulnerability profiles. Model-level evaluation reveals baseline differences: GPT-OSS-20B (39.47\% ASR) versus Gemini-2.0-flash (50.00\% ASR), with both models showing susceptibility to social engineering. However, agentic-level assessment exposes agent-specific risks invisible to traditional evaluation. We discover "agentic-only" vulnerabilities that emerge exclusively in agentic contexts, with tool-calling showing 24-60\% higher ASR across both models. Cross-model analysis reveals universal agentic patterns, where agent transfer operations as highest-risk tools, with semantic pattern revealed rather than syntactic vulnerability mechanisms. Direct attack transfer from model-level to agentic contexts shows degraded performance of successful prompts (GPT-OSS-20B: 57\% human injection ASR; Gemini-2.0-flash: 28\%), while context-aware iterative attacks successfully compromise objectives that failed at model-level, confirming systematic vulnerabilities gaps. Action-based prompt improvement substantially reduces action-averaged agentic jailbreak success on GPT-OSS-20B (direct: 45.3\%

Reinforcement Learning has emerged as a dominant post-training approach to elicit agentic RAG behaviors such as search and planning from language models. Despite its success with larger models, applying RL to compact models (e.g., 0.5--1B parameters) presents unique challenges. The compact models exhibit poor initial performance, resulting in sparse rewards and unstable training. To overcome these difficulties, we propose Distillation-Guided Policy Optimization (DGPO), which employs cold-start initialization from teacher demonstrations and continuous teacher guidance during policy optimization. To understand how compact models preserve agentic behavior, we introduce Agentic RAG Capabilities (ARC), a fine-grained metric analyzing reasoning, search coordination, and response synthesis. Comprehensive experiments demonstrate that DGPO enables compact models to achieve sophisticated agentic search behaviors, even outperforming the larger teacher model in some cases. DGPO makes agentic RAG feasible in computing resource-constrained environments.

Single-Pixel Imaging (SPI) enables the reconstruction of objects using a single detector through sequential illuminations with structured light patterns. The choice of illumination patterns is critical, particularly in highly undersampled regimes, where it directly determines reconstruction quality and acquisition speed. Instead of relying on handcrafted or fixed patterns, we propose to learn task-specific patterns directly from data. Practical SPI hardware only supports binary patterns, making binary pattern design a necessary consideration. We propose a bilevel optimisation method for learning task-specific binary illumination patterns optimised for applications such as single-pixel fluorescence microscopy. We address the non-differentiable nature of binary optimisation using the Straight-Through Estimator. In addition, we incorporate learned variational regularisation, improving reconstruction quality and robustness. We demonstrate our method on the CytoImageNet microscopy dataset. We show that our learned patterns achieve superior reconstruction performance compared to baseline methods and end-to-end deep learning, particularly in highly undersampled regimes and in scarce-data settings.

As large language models (LLMs) are increasingly deployed in real-world applications, the need to selectively remove unwanted knowledge while preserving model utility has become paramount. Recent work has explored sparse autoencoders (SAEs) to perform precise interventions on monosemantic features. However, most SAE-based methods operate at inference time, which does not create persistent changes in the model's parameters. Such interventions can be bypassed or reversed by malicious actors with parameter access. We introduce CRISP, a parameter-efficient method for persistent concept unlearning using SAEs. CRISP automatically identifies salient SAE features across multiple layers and suppresses their activations. We experiment with two LLMs and show that our method outperforms prior approaches on safety-critical unlearning tasks from the WMDP benchmark, successfully removing harmful knowledge while preserving general and in-domain capabilities. Feature-level analysis reveals that CRISP achieves semantically coherent separation between target and benign concepts, allowing precise suppression of the target features.

Membership inference attacks serves as useful tool for fair use of language models, such as detecting potential copyright infringement and auditing data leakage. However, many current state-of-the-art attacks require access to models' hidden states or probability distribution, which prevents investigation into more widely-used, API-access only models like GPT-4. In this work, we introduce N-Gram Coverage Attack, a membership inference attack that relies solely on text outputs from the target model, enabling attacks on completely black-box models. We leverage the observation that models are more likely to memorize and subsequently generate text patterns that were commonly observed in their training data. Specifically, to make a prediction on a candidate member, N-Gram Coverage Attack first obtains multiple model generations conditioned on a prefix of the candidate. It then uses n-gram overlap metrics to compute and aggregate the similarities of these outputs with the ground truth suffix; high similarities indicate likely membership. We first demonstrate on a diverse set of existing benchmarks that N-Gram Coverage Attack outperforms other black-box methods while also impressively achieving comparable or even better performance to state-of-the-art white-box attacks - despite having access to only text outputs. Interestingly, we find that the success rate of our method scales with the attack compute budget - as we increase the number of sequences generated from the target model conditioned on the prefix, attack performance tends to improve. Having verified the accuracy of our method, we use it to investigate previously unstudied closed OpenAI models on multiple domains. We find that more recent models, such as GPT-4o, exhibit increased robustness to membership inference, suggesting an evolving trend toward improved privacy protections.

Retrieval-Augmented Generation (RAG) has emerged as an effective paradigm for expanding the knowledge capacity of Multimodal Large Language Models (MLLMs) by incorporating external knowledge sources into the generation process, and has been widely adopted for knowledge-based Visual Question Answering (VQA). Despite impressive advancements, vanilla RAG-based VQA methods that rely on unstructured documents and overlook the structural relations among knowledge elements frequently introduce irrelevant or misleading content, degrading answer accuracy and reliability. To overcome these challenges, a promising solution is to integrate multimodal knowledge graphs (KGs) into RAG-based VQA frameworks, thereby enhancing generation through structured multimodal knowledge. To this end, this paper proposes mKG-RAG, a novel retrieval-augmented generation framework built upon multimodal KGs for knowledge-intensive VQA tasks. Specifically, mKG-RAG leverages MLLM-driven graph extraction and vision-text matching to distill semantically consistent, modality-complementary entities and relations from multimodal documents, constructing high-quality multimodal KGs as structured knowledge representations. Furthermore, a dual-stage retrieval strategy equipped with a query-aware multimodal retriever is introduced to improve retrieval efficiency while progressively refining precision. Comprehensive experiments demonstrate that our approach significantly outperforms existing approaches and sets new state-of-the-art results for knowledge-based VQA. The code is available at https://github.com/xandery-geek/mKG-RAG.

Planning collision-free paths for a large group of agents is a challenging problem in many real-world applications. While recent advances in Multi-Agent Path Finding (MAPF) have shown promising progress, standard MAPF planners continue to rely on simplified kinodynamic models, preventing agents from directly following the generated MAPF plan. To bridge this gap, we propose kinodynamic Temporal Plan Graph planning (kTPG), a multi-agent speed optimization algorithm that efficiently refines a MAPF plan into a set of kinodynamically feasible speed profiles. We further incorporate execution timing uncertainty models and provide deterministic guarantees under bounded uncertainty models and probabilistic guarantees under stochastic models. Building on kTPG, we propose Windowed kTPG (WinkTPG), a MAPF execution framework that incrementally refines MAPF plans using a window-based mechanism, dynamically incorporating agent information during execution to reduce uncertainty. Experiments show that WinkTPG can generate speed profiles for up to 1,000 agents within 1 second and improves solution quality by up to 51.7% over existing MAPF execution methods. We further validate WinkTPG in high-fidelity physics-based simulation and on real-world robots.

Sea surface temperature (SST) prediction is a critical task in ocean science, supporting various applications, such as weather forecasting, fisheries management, and storm tracking. While existing data-driven methods have demonstrated significant success, they often neglect to leverage the rich domain knowledge accumulated over the past decades, limiting further advancements in prediction accuracy. The recent emergence of large language models (LLMs) has highlighted the potential of integrating domain knowledge for downstream tasks. However, the application of LLMs to SST prediction remains underexplored, primarily due to the challenge of integrating ocean domain knowledge and numerical data. To address this issue, we propose Ocean Knowledge Graph-enhanced LLM (OKG-LLM), a novel framework for global SST prediction. To the best of our knowledge, this work presents the first systematic effort to construct an Ocean Knowledge Graph (OKG) specifically designed to represent diverse ocean knowledge for SST prediction. We then develop a graph embedding network to learn the comprehensive semantic and structural knowledge within the OKG, capturing both the unique characteristics of individual sea regions and the complex correlations between them. Finally, we align and fuse the learned knowledge with fine-grained numerical SST data and leverage a pre-trained LLM to model SST patterns for accurate prediction. Extensive experiments on the real-world dataset demonstrate that OKG-LLM consistently outperforms state-of-the-art methods, showcasing its effectiveness, robustness, and potential to advance SST prediction. The codes are available in the online repository.

In the oil industry, undesirable events in oil wells can cause economic losses, environmental accidents, and human casualties. Solutions based on Artificial Intelligence and Machine Learning for Early Detection of such events have proven valuable for diverse applications across industries. In 2019, recognizing the importance and the lack of public datasets related to undesirable events in oil wells, Petrobras developed and publicly released the first version of the 3W Dataset, which is essentially a set of Multivariate Time Series labeled by experts. Since then, the 3W Dataset has been developed collaboratively and has become a foundational reference for numerous works in the field. This data article describes the current publicly available version of the 3W Dataset, which contains structural modifications and additional labeled data. The detailed description provided encourages and supports the 3W community and new 3W users to improve previous published results and to develop new robust methodologies, digital products and services capable of detecting undesirable events in oil wells with enough anticipation to enable corrective or mitigating actions.

Estimating single-cell responses across various perturbations facilitates the identification of key genes and enhances drug screening, significantly boosting experimental efficiency. However, single-cell sequencing is a destructive process, making it impossible to capture the same cell's phenotype before and after perturbation. Consequently, data collected under perturbed and unperturbed conditions are inherently unpaired, creating a critical yet unresolved problem in single-cell perturbation modeling. Moreover, the high dimensionality and sparsity of single-cell expression make direct modeling prone to focusing on zeros and neglecting meaningful patterns. To address these problems, we propose a new paradigm for single-cell perturbation modeling. Specifically, we leverage dual diffusion models to learn the control and perturbed distributions separately, and implicitly align them through a shared Gaussian latent space, without requiring explicit cell pairing. Furthermore, we introduce a sparsity masking strategy in which the mask model learns to predict zero-expressed genes, allowing the diffusion model to focus on capturing meaningful patterns among expressed genes and thereby preserving diversity in high-dimensional sparse data. We introduce \textbf{Doloris}, a generative framework that defines a new paradigm for modeling unpaired, high-dimensional, and sparse single-cell perturbation data. It leverages dual conditional diffusion models for separate learning of control and perturbed distributions, complemented by a sparsity masking strategy to enhance prediction of zero-valued genes. The results on publicly available datasets show that our model effectively captures the diversity of single-cell perturbations and achieves state-of-the-art performance. To facilitate reproducibility, we include the code in the supplementary materials.

Ensuring the safety and alignment of Large Language Models is a significant challenge with their growing integration into critical applications and societal functions. While prior research has primarily focused on jailbreak attacks, less attention has been given to non-adversarial failures that subtly emerge during benign interactions. We introduce secondary risks a novel class of failure modes marked by harmful or misleading behaviors during benign prompts. Unlike adversarial attacks, these risks stem from imperfect generalization and often evade standard safety mechanisms. To enable systematic evaluation, we introduce two risk primitives verbose response and speculative advice that capture the core failure patterns. Building on these definitions, we propose SecLens, a black-box, multi-objective search framework that efficiently elicits secondary risk behaviors by optimizing task relevance, risk activation, and linguistic plausibility. To support reproducible evaluation, we release SecRiskBench, a benchmark dataset of 650 prompts covering eight diverse real-world risk categories. Experimental results from extensive evaluations on 16 popular models demonstrate that secondary risks are widespread, transferable across models, and modality independent, emphasizing the urgent need for enhanced safety mechanisms to address benign yet harmful LLM behaviors in real-world deployments.

Neural Processes (NPs) are a rapidly evolving class of models designed to directly model the posterior predictive distribution of stochastic processes. While early architectures were developed primarily as a scalable alternative to Gaussian Processes (GPs), modern NPs tackle far more complex and data-hungry applications spanning geology, epidemiology, climate, and robotics. These applications have placed increasing pressure on the scalability of these models, with many architectures compromising accuracy for scalability. In this paper, we demonstrate that this trade-off is often unnecessary, particularly when modeling fully or partially translation-invariant processes. We propose a versatile new architecture, the Biased Scan Attention Transformer Neural Process (BSA-TNP), which introduces Kernel Regression Blocks (KRBlocks), group-invariant attention biases, and memory-efficient Biased Scan Attention (BSA). BSA-TNP is able to: (1) match or exceed the accuracy of the best models while often training in a fraction of the time, (2) exhibit translation invariance, enabling learning at multiple resolutions simultaneously, (3) transparently model processes that evolve in both space and time, (4) support high-dimensional fixed effects, and (5) scale gracefully, running inference on over 1M test points and 100K context points in under a minute on a single 24GB GPU. Code is provided as part of the `dl4bi` package.

We propose Guided Speculative Inference (GSI), a novel algorithm for efficient reward-guided decoding in large language models. GSI combines soft best-of-$n$ test-time scaling with a reward model $r(x,y)$ and speculative samples from a small auxiliary model $π_S(y\mid x)$. We provably approximate both the optimal tilted policy $π_{β,B}(y\mid x) \propto π_B(y\mid x)\exp(β\,r(x,y))$ of soft best-of-$n$ under the base model $π_B$, as well as the expected reward under the optimal policy. In experiments on reasoning benchmarks (MATH500, OlympiadBench, Minerva Math, MMLU-STEM, GSM8K) and across different model families, our method achieves higher accuracy than standard soft best-of-$n$ with $π_S$ and reward-guided speculative decoding (Liao et al., 2025), and in certain settings even outperforms soft best-of-$n$ with $π_B$, while reducing end-to-end latency by up to $28\%$. The code is available at https://github.com/j-geuter/GSI .

Robotic surgery represents a major breakthrough in medical interventions, which has revolutionized surgical procedures. However, the high cost and limited accessibility of robotic surgery systems pose significant challenges for training purposes. This study addresses these issues by developing a cost-effective robotic laparoscopy training system that closely replicates advanced robotic surgery setups to ensure broad access for both on-site and remote users. Key innovations include the design of a low-cost robotic end-effector that effectively mimics high-end laparoscopic instruments. Additionally, a digital twin platform was established, facilitating detailed simulation, testing, and real-time monitoring, which enhances both system development and deployment. Furthermore, teleoperation control was optimized, leading to improved trajectory tracking while maintaining remote center of motion (RCM) constraint, with a RMSE of 5 μm and reduced system latency to 0.01 seconds. As a result, the system provides smooth, continuous motion and incorporates essential safety features, making it a highly effective tool for laparoscopic training.

Text-to-image retrieval (T2I retrieval) remains challenging because cross-modal embeddings often behave as bags of concepts, underrepresenting structured visual relationships such as pose and viewpoint. We proposeVisualize-then-Retrieve (VisRet), a retrieval paradigm that mitigates this limitation of cross-modal similarity alignment. VisRet first projects textual queries into the image modality via T2I generation, then performs retrieval within the image modality to bypass the weaknesses of cross-modal retrievers in recognizing subtle visual-spatial features. Across four benchmarks (Visual-RAG, INQUIRE-Rerank, Microsoft COCO, and our new Visual-RAG-ME featuring multi-entity comparisons), VisRet substantially outperforms cross-modal similarity matching and baselines that recast T2I retrieval as text-to-text similarity matching, improving nDCG@30 by 0.125 on average with CLIP as the retriever and by 0.121 with E5-V. For downstream question answering, VisRet increases accuracy on Visual-RAG and Visual-RAG-ME by 3.8% and 15.7% in top-1 retrieval, and by 3.9% and 11.1% in top-10 retrieval. Ablation studies show compatibility with different T2I instruction LLMs, T2I generation models, and downstream LLMs. VisRet provides a simple yet effective perspective for advancing in text-image retrieval. Our code and the new benchmark are publicly available at https://github.com/xiaowu0162/Visualize-then-Retrieve.

The seminal breakthrough of AlphaFold in protein structure prediction relied on a learned potential energy function parameterized by deep models, in contrast to its successors AlphaFold2 and AlphaFold3, which lack an explicit probabilistic interpretation. While AlphaFold's potential was originally justified by heuristic analogy to physical potentials of mean force, we show that it can instead be understood as a principled instance of probability kinematics (PK), also known as Jeffrey conditioning, a generalization of Bayesian updating. This reinterpretation reveals that AlphaFold is a generalized Bayesian model that explicitly defines a posterior distribution over structures, providing a deeper explanation of its success and a foundation for future model design. To demonstrate this framework with precision, we introduce a tractable synthetic model in which an angular random walk prior is updated with distance-based evidence via PK, directly mirroring AlphaFold's mechanism. This setting allows us to explore the probabilistic foundations of AlphaFold in a clear and interpretable way. Our work connects a landmark in protein structure prediction to a broader class of compositional deep generative models and points to new opportunities for principled probabilistic approaches.

Understanding sources of a model's uncertainty regarding its predictions is crucial for effective human-AI collaboration. Prior work proposes using numerical uncertainty or hedges ("I'm not sure, but ..."), which do not explain uncertainty that arises from conflicting evidence, leaving users unable to resolve disagreements or rely on the output. We introduce CLUE (Conflict-and-Agreement-aware Language-model Uncertainty Explanations), the first framework to generate natural language explanations of model uncertainty by (i) identifying relationships between spans of text that expose claim-evidence or inter-evidence conflicts and agreements that drive the model's predictive uncertainty in an unsupervised way, and (ii) generating explanations via prompting and attention steering that verbalize these critical interactions. Across three language models and two fact-checking datasets, we show that CLUE produces explanations that are more faithful to the model's uncertainty and more consistent with fact-checking decisions than prompting for uncertainty explanations without span-interaction guidance. Human evaluators judge our explanations to be more helpful, more informative, less redundant, and more logically consistent with the input than this baseline. CLUE requires no fine-tuning or architectural changes, making it plug-and-play for any white-box language model. By explicitly linking uncertainty to evidence conflicts, it offers practical support for fact-checking and generalises readily to other tasks that require reasoning over complex information.

This work aims at a challenging task: human action-reaction synthesis, i.e., generating human reactions conditioned on the action sequence of another person. Currently, autoregressive modeling approaches with vector quantization (VQ) have achieved remarkable performance in motion generation tasks. However, VQ has inherent disadvantages, including quantization information loss, low codebook utilization, etc. In addition, while dividing the body into separate units can be beneficial, the computational complexity needs to be considered. Also, the importance of mutual perception among units is often neglected. In this work, we propose MARRS, a novel framework designed to generate coordinated and fine-grained reaction motions using continuous representations. Initially, we present the Unit-distinguished Motion Variational AutoEncoder (UD-VAE), which segments the entire body into distinct body and hand units, encoding each independently. Subsequently, we propose Action-Conditioned Fusion (ACF), which involves randomly masking a subset of reactive tokens and extracting specific information about the body and hands from the active tokens. Furthermore, we introduce Mutual Unit Modulation (MUM) to facilitate interaction between body and hand units by using the information from one unit to adaptively modulate the other. Finally, for the diffusion model, we employ a compact MLP as a noise predictor for each distinct body unit and incorporate the diffusion loss to model the probability distribution of each token. Both quantitative and qualitative results demonstrate that our method achieves superior performance. Project page: https://aigc-explorer.github.io/MARRS/.

Models and methods originally developed for Novel View Synthesis and Scene Rendering, such as Neural Radiance Fields (NeRF) and Gaussian Splatting, are increasingly being adopted as representations in Simultaneous Localization and Mapping (SLAM). However, existing datasets fail to include the specific challenges of both fields, such as sequential operations and, in many settings, multi-modality in SLAM or generalization across viewpoints and illumination conditions in neural rendering. Additionally, the data are often collected using sensors which are handheld or mounted on drones or mobile robots, which complicates the accurate reproduction of sensor motions. To bridge these gaps, we introduce SLAM&Render, a novel dataset designed to benchmark methods in the intersection between SLAM, Novel View Rendering and Gaussian Splatting. Recorded with a robot manipulator, it uniquely includes 40 sequences with time-synchronized RGB-D images, IMU readings, robot kinematic data, and ground-truth pose streams. By releasing robot kinematic data, the dataset also enables the assessment of recent integrations of SLAM paradigms within robotic applications. The dataset features five setups with consumer and industrial objects under four controlled lighting conditions, each with separate training and test trajectories. All sequences are static with different levels of object rearrangements and occlusions. Our experimental results, obtained with several baselines from the literature, validate SLAM&Render as a relevant benchmark for this emerging research area.

Foundation Models, which leverage large neural networks pre-trained on unlabelled data before fine-tuning for specific tasks, are increasingly being applied to specialised domains. Recent examples include ClimaX for climate and Clay for satellite Earth observation, but a Foundation Model for Space Object Behavioural Analysis has not yet been developed. As orbital populations grow, automated methods for characterising space object behaviour are crucial for space safety. Here, we present a self-supervised framework for space object behavioural analysis, representing a first step towards a Foundation Model for SOBA. The backbone is a Perceiver-Variational Autoencoder (VAE) architecture, pre-trained with self-supervised reconstruction and masked reconstruction on 227,000 light curves from the MMT-9 observatory. The VAE enables anomaly detection, motion prediction, and synthetic light curve generation. We fine-tuned the model using two independent light curve simulators (CASSANDRA and GRIAL), with CAD models of boxwing, Sentinel-3, SMOS, and Starlink platforms. Our pre-trained model achieved a reconstruction mean squared error of 0.009, identifying potentially anomalous light curves through reconstruction difficulty. After fine-tuning, the model scored 85% and 82% accuracy, with 0.92 and 0.95 ROC AUC scores in anomaly detection and motion mode prediction (e.g., sun-pointing, spin, tumbling). Analysis of high-confidence predictions on real data revealed distinct patterns including characteristic object profiles and satellite glinting. Our work demonstrates how self-supervised learning can simultaneously enable anomaly detection, motion prediction, and synthetic data generation from rich pre-trained representations, supporting space safety and sustainability through automated monitoring and simulation.

Uniform manifold approximation and projection (UMAP) is among the most popular neighbor embedding methods. The method samples pairs of point indices according to similarities in the high-dimensional space, and applies attractive and repulsive forces to their coordinates in the low-dimensional embedding. In this paper, we analyze the forces to reveal their effects on cluster formations and visualization, and compare UMAP to its contemporaries. Repulsion emphasizes differences, controlling cluster boundaries and inter-cluster distance. Attraction is more subtle, as attractive tension between points can manifest simultaneously as attraction and repulsion in the lower-dimensional mapping. This explains the need for learning rate annealing and motivates the different treatments between attractive and repulsive terms. Moreover, by modifying attraction, we improve the consistency of cluster formation under random initialization. Overall, our analysis provides a mechanistic understanding of UMAP and related embedding methods.

Open-ended learning agents must efficiently prioritize goals in vast possibility spaces, focusing on those that maximize learning progress (LP). When such autotelic exploration is achieved by LLM agents trained with online RL in high-dimensional and evolving goal spaces, a key challenge for LP prediction is modeling one's own competence, a form of metacognitive monitoring. Traditional approaches either require extensive sampling or rely on brittle expert-defined goal groupings. We introduce MAGELLAN, a metacognitive framework that lets LLM agents learn to predict their competence and LP online. By capturing semantic relationships between goals, MAGELLAN enables sample-efficient LP estimation and dynamic adaptation to evolving goal spaces through generalization. In an interactive learning environment, we show that MAGELLAN improves LP prediction efficiency and goal prioritization, being the only method allowing the agent to fully master a large and evolving goal space. These results demonstrate how augmenting LLM agents with a metacognitive ability for LP predictions can effectively scale curriculum learning to open-ended goal spaces.

Large Language Models (LLMs) have made significant strides in code generation and problem solving. Current approaches employ external tool-based iterative debuggers that use compiler or other tool-based runtime feedback to refine coarse programs generated by various methods. However, the effectiveness of these approaches heavily relies on the quality of the initial code generation, which remains an open challenge. In this paper, we introduce CodeSim, a novel multi-agent code generation framework that comprehensively addresses the stages of program synthesis-planning, coding, and debugging-through a human-like perception approach. As human verifies their understanding of any algorithms through visual simulation, CodeSim uniquely features a method of plan verification and internal debugging through the step-by-step simulation of input/output. Extensive experiments across seven challenging competitive problem-solving and program synthesis benchmarks demonstrate CodeSim's remarkable code generation capabilities. Our framework achieves new state-of-the-art (pass@1) results-(HumanEval 95.1%, MBPP 90.7%, APPS 22%, and CodeContests 29.1%). Furthermore, our method shows potential for even greater enhancement when cascaded with external debuggers. To facilitate further research and development in this area, we have open-sourced our framework in this link (https://kagnlp.github.io/codesim.github.io/).

With the integration of multimodal large language models (MLLMs) into robotic systems and AI applications, embedding emotional intelligence (EI) capabilities is essential for enabling these models to perceive, interpret, and respond to human emotions effectively in real-world scenarios. Existing static, text-based, or text-image benchmarks overlook the multimodal complexities of real interactions and fail to capture the dynamic, context-dependent nature of emotional expressions, rendering them inadequate for evaluating MLLMs' EI capabilities. To address these limitations, we introduce EmoBench-M, a systematic benchmark grounded in established psychological theories, designed to evaluate MLLMs across 13 evaluation scenarios spanning three hierarchical dimensions: foundational emotion recognition (FER), conversational emotion understanding (CEU), and socially complex emotion analysis (SCEA). Evaluation was conducted on 27 state-of-the-art MLLMs, using both objective task-specific metrics and LLM-based evaluation, revealing a substantial performance gap relative to human-level competence. Even the best performing models, Gemini-3.0-Pro and GPT-5.2, achieve the highest scores on EmoBench-M, 70.5 and 66.5 points respectively. Specialized models such as AffectGPT exhibit uneven performance across EmoBench-M, demonstrating strengths in certain scenarios but generally lacking comprehensive emotional intelligence. By providing a comprehensive, multimodal evaluation framework, EmoBench-M captures both the strengths and weaknesses of current MLLMs across diverse emotional contexts. All benchmark resources, including datasets and code, are publicly available at https://emo-gml.github.io/, facilitating further research and advancement in MLLM emotional intelligence.

End-to-end representation learning has become a powerful tool for estimating causal quantities from high-dimensional observational data, but its efficiency remained unclear. Here, we face a central tension: End-to-end representation learning methods often work well in practice but lack asymptotic optimality in the form of the quasi-oracle efficiency. In contrast, two-stage Neyman-orthogonal learners provide such a theoretical optimality property but do not explicitly benefit from the strengths of representation learning. In this work, we step back and ask two research questions: (1) When do representations strengthen existing Neyman-orthogonal learners? and (2) Can a balancing constraint - a commonly proposed technique in the representation learning literature - provide improvements to Neyman-orthogonality? We address these two questions through our theoretical and empirical analysis, where we introduce a unifying framework that connects representation learning with Neyman-orthogonal learners (namely, OR-learners). In particular, we show that, under the low-dimensional manifold hypothesis, the OR-learners can strictly improve the estimation error of the standard Neyman-orthogonal learners. At the same time, we find that the balancing constraint requires an additional inductive bias and cannot generally compensate for the lack of Neyman-orthogonality of the end-to-end approaches. Building on these insights, we offer guidelines for how users can effectively combine representation learning with the classical Neyman-orthogonal learners to achieve both practical performance and theoretical guarantees.

In the field of deep learning-based computer vision, YOLO is revolutionary. With respect to deep learning models, YOLO is also the one that is evolving the most rapidly. Unfortunately, not every YOLO model possesses scholarly publications. Moreover, there exists a YOLO model that lacks a publicly accessible official architectural diagram. Naturally, this engenders challenges, such as complicating the understanding of how the model operates in practice. Furthermore, the review articles that are presently available do not investigate the specifics of each model. The objective of this study is to present a comprehensive and in-depth architecture comparison of the four most recent YOLO models, specifically YOLOv8 through YOLO11, thereby enabling readers to quickly grasp not only how each model functions, but also the distinctions between them. To analyze each YOLO version's architecture, we meticulously examined the relevant academic papers, documentation, and scrutinized the source code. The analysis reveals that while each version of YOLO has improvements in architecture and feature extraction, certain blocks remain unchanged. The lack of scholarly publications and official diagrams presents challenges for understanding the model's functionality and future enhancement. Future developers are encouraged to provide these resources.

Aiming to accelerate the training of large deep neural networks (DNN) in an energy-efficient way, analog in-memory computing (AIMC) emerges as a solution with immense potential. AIMC accelerator keeps model weights in memory without moving them from memory to processors during training, reducing overhead dramatically. Despite its efficiency, scaling up AIMC systems presents significant challenges. Since weight copying is expensive and inaccurate, data parallelism is less efficient on AIMC accelerators. It necessitates the exploration of pipeline parallelism, particularly asynchronous pipeline parallelism, which utilizes all available accelerators during the training process. This paper examines the convergence theory of stochastic gradient descent on AIMC hardware with an asynchronous pipeline (Analog-SGD-AP). Although there is empirical exploration of AIMC accelerators, the theoretical understanding of how analog hardware imperfections in weight updates affect the training of multi-layer DNN models remains underexplored. Furthermore, the asynchronous pipeline parallelism results in stale weights issues, which render the update signals no longer valid gradients. To close the gap, this paper investigates the convergence properties of Analog-SGD-AP on multi-layer DNN training. We show that the Analog-SGD-AP converges with iteration complexity $O(\varepsilon^{-2}+\varepsilon^{-1})$ despite the aforementioned issues, which matches the complexities of digital SGD and Analog SGD with synchronous pipeline, except the non-dominant term $O(\varepsilon^{-1})$. It implies that AIMC training benefits from asynchronous pipelining almost for free compared with the synchronous pipeline by overlapping computation.

Medical palpation is more than force transmission. It is a bidirectional cognitive and emotional exchange between doctor and patient. We model two complementary touch pathways: active touch by the doctor (kinesthetic and tactile) and passive touch by the patient (subjective and emotional). We use this framework to design a mixed-reality telepalpation prototype and evaluate it with 14 experienced clinicians serving as both doctors and patients across 391 trials. Touch location was transmitted reliably across participants, while force perception showed systematic inter-individual variation, suggesting that force alone is insufficient to characterize the palpation experience.

Large Vision Language Models (LVLMs) are increasingly integral to healthcare applications, including medical visual question answering and imaging report generation. While these models inherit the robust capabilities of foundational Large Language Models (LLMs), they also inherit susceptibility to hallucinations-a significant concern in high-stakes medical contexts where the margin for error is minimal. However, currently, there are no dedicated methods or benchmarks for hallucination detection and evaluation in the medical field. To bridge this gap, we introduce Med-HallMark, the first benchmark specifically designed for hallucination detection and evaluation within the medical multimodal domain. This benchmark provides multi-tasking hallucination support, multifaceted hallucination data, and hierarchical hallucination categorization. Furthermore, we propose the MediHall Score, a new medical evaluative metric designed to assess LVLMs' hallucinations through a hierarchical scoring system that considers the severity and type of hallucination, thereby enabling a granular assessment of potential clinical impacts. We also present MediHallDetector, a novel Medical LVLM engineered for precise hallucination detection, which employs multitask training for hallucination detection. Through extensive experimental evaluations, we establish baselines for popular LVLMs using our benchmark. The findings indicate that MediHall Score provides a more nuanced understanding of hallucination impacts compared to traditional metrics and demonstrate the enhanced performance of MediHallDetector. We hope this work can significantly improve the reliability of LVLMs in medical applications. All resources of this work have been released at https://github.com/ydk122024/Med-HallMark.

Mental health constitutes a complex and pervasive global challenge, affecting millions of lives and often leading to severe consequences. In this paper, we conduct a thorough survey to explore the intersection of data science, artificial intelligence, and mental healthcare, focusing on the recent developments of mental disorder detection through online social media (OSM). A significant portion of the population actively engages in OSM platforms, creating a vast repository of personal data that holds immense potential for mental health analytics. The paper navigates through traditional diagnostic methods, state-of-the-art data- and AI-driven research studies, and the emergence of explainable AI (XAI) models for mental healthcare. We review state-of-the-art machine learning methods, particularly those based on modern deep learning, while emphasising the need for explainability in healthcare AI models. The experimental design section provides insights into prevalent practices, including available datasets and evaluation approaches. We also identify key issues and challenges in the field and propose promising future research directions. As mental health decisions demand transparency, interpretability, and ethical considerations, this paper contributes to the ongoing discourse on advancing XAI in mental healthcare through social media. The comprehensive overview presented here aims to guide researchers, practitioners, and policymakers in developing the area of mental disorder detection.