Large Vision-Language Models (LVLMs) typically process visual inputs as a prefix to the language decoder. As the model autoregressively generates text, this initial visual information inevitably undergoes "dilution" leading the model to over-rely on language priors and hallucinate objects. Existing interventions attempt to correct this by contrasting logits or iteratively refining outputs, but they incur prohibitive latency costs. We propose Residual-Update Directed DEcoding Regulation (RUDDER), a framework that counters visual dilution by creating a persistent visual anchor. We extract a robust evidence direction (CARD) directly from the model's prefill residual updates, and inject it into the decoding process. This injection is modulated by an adaptive gate, the Beta Gate, which acts as a trust mechanism and ensures the visual reminder is applied only when necessary. Experiments on LLaVA-1.5 (7B/13B), Idefics2, InstructBLIP, and Qwen2.5-VL demonstrate that RUDDER consistently mitigates hallucination (with greedy decoding, RUDDER reduces CHAIR_S by an average of 24.4% and CHAIR_i by 23.6% relative) and scales effectively across architectures, all while maintaining >96.0% throughput.

Global plant maps of plant traits, such as leaf nitrogen or plant height, are essential for understanding ecosystem processes, including the carbon and energy cycles of the Earth system. However, existing trait maps remain limited by the high cost and sparse geographic coverage of field-based measurements. Citizen science initiatives offer a largely untapped resource to overcome these limitations, with over 50 million geotagged plant photographs worldwide capturing valuable visual information on plant morphology and physiology. In this study, we introduce PlantTraitNet, a multi-modal, multi-task uncertainty-aware deep learning framework that predictsfour key plant traits (plant height, leaf area, specific leaf area, and nitrogen content) from citizen science photos using weak supervision. By aggregating individual trait predictions across space, we generate global maps of trait distributions. We validate these maps against independent vegetation survey data (sPlotOpen) and benchmark them against leading global trait products. Our results show that PlantTraitNet consistently outperforms existing trait maps across all evaluated traits, demonstrating that citizen science imagery, when integrated with computer vision and geospatial AI, enables not only scalable but also more accurate global trait mapping. This approach offers a powerful new pathway for ecological research and Earth system modeling.

Short-video recommenders such as Douyin must exploit extremely long user behavior histories without breaking latency or cost budgets. We present an end-to-end industrial recommender system that scales long-sequence recommendation modeling to 10K-length histories in production. First, we introduce Stacked Target-to-History Cross Attention (STCA), which replaces history self-attention with stacked cross-attention from the target to the history, reducing complexity from quadratic to linear in sequence length and enabling efficient end-to-end training over long user behavior sequences. Second, we propose Request Level Batching (RLB), a user-centric batching scheme that aggregates multiple targets for the same user/request to share the user-side encoding, substantially lowering sequence-related storage, communication, and compute without changing the learning objective. Third, we design a length-extrapolative training strategy -- train on shorter windows, infer on much longer ones -- so the model generalizes to 10K-scale histories without additional training cost. Across offline and online experiments, we observe predictable, monotonic gains as we scale history length and model capacity, mirroring the scaling law behavior observed in large language models. Deployed at full traffic on Douyin, our system delivers significant improvements on key engagement metrics while meeting production latency, demonstrating a practical path to scaling end-to-end ultra-long sequence recommendation to the 10K regime.

Multiple-choice cloze questions are commonly used to assess linguistic proficiency and comprehension. However, generating high-quality distractors remains challenging, as existing methods often lack adaptability and control over difficulty levels, and the absence of difficulty-annotated datasets further hinders progress. To address these issues, we propose a novel framework for generating distractors with controllable difficulty by leveraging both data augmentation and a multitask learning strategy. First, to create a high-quality, difficulty-annotated dataset, we introduce a two-way distractor generation process to produce diverse and plausible distractors. These candidates are filtered and then categorized by difficulty using an ensemble QA system. Second, this newly created dataset is used to train a difficulty-controllable generation model via multitask learning. Experimental results demonstrate that our method generates high-quality distractors across difficulty levels and substantially outperforms GPT-4o in aligning distractor difficulty with human perception.

Online bilevel optimization (OBO) is a powerful framework for machine learning problems where both outer and inner objectives evolve over time, requiring dynamic updates. Current OBO approaches rely on deterministic \textit{window-smoothed} regret minimization, which may not accurately reflect system performance when functions change rapidly. In this work, we introduce a novel search direction and show that both first- and zeroth-order (ZO) stochastic OBO algorithms leveraging this direction achieve sublinear {stochastic bilevel regret without window smoothing}. Beyond these guarantees, our framework enhances efficiency by: (i) reducing oracle dependence in hypergradient estimation, (ii) updating inner and outer variables alongside the linear system solution, and (iii) employing ZO-based estimation of Hessians, Jacobians, and gradients. Experiments on online parametric loss tuning and black-box adversarial attacks validate our approach.

Estimating the cognitive complexity of reading comprehension (RC) items is crucial for assessing item difficulty before it is administered to learners. Unlike syntactic and semantic features, such as passage length or semantic similarity between options, cognitive features that arise during answer reasoning are not readily extractable using existing NLP tools and have traditionally relied on human annotation. In this study, we examine whether large language models (LLMs) can estimate the cognitive complexity of RC items by focusing on two dimensions-Evidence Scope and Transformation Level-that indicate the degree of cognitive burden involved in reasoning about the answer. Our experimental results demonstrate that LLMs can approximate the cognitive complexity of items, indicating their potential as tools for prior difficulty analysis. Further analysis reveals a gap between LLMs' reasoning ability and their metacognitive awareness: even when they produce correct answers, they sometimes fail to correctly identify the features underlying their own reasoning process.

Fair graph clustering seeks partitions that respect network structure while maintaining proportional representation across sensitive groups, with applications spanning community detection, team formation, resource allocation, and social network analysis. Many existing approaches enforce rigid constraints or rely on multi-stage pipelines (e.g., spectral embedding followed by $k$-means), limiting trade-off control, interpretability, and scalability. We introduce \emph{DFNMF}, an end-to-end deep nonnegative tri-factorization tailored to graphs that directly optimizes cluster assignments with a soft statistical-parity regularizer. A single parameter $λ$ tunes the fairness--utility balance, while nonnegativity yields parts-based factors and transparent soft memberships. The optimization uses sparse-friendly alternating updates and scales near-linearly with the number of edges. Across synthetic and real networks, DFNMF achieves substantially higher group balance at comparable modularity, often dominating state-of-the-art baselines on the Pareto front. The code is available at https://github.com/SiamakGhodsi/DFNMF.git.

Deep learning models have achieved remarkable accuracy in chest X-ray diagnosis, yet their widespread clinical adoption remains limited by the black-box nature of their predictions. Clinicians require transparent, verifiable explanations to trust automated diagnoses and identify potential failure modes. We introduce CXR-LanIC (Language-Grounded Interpretable Classifier for Chest X-rays), a novel framework that addresses this interpretability challenge through task-aligned pattern discovery. Our approach trains transcoder-based sparse autoencoders on a BiomedCLIP diagnostic classifier to decompose medical image representations into interpretable visual patterns. By training an ensemble of 100 transcoders on multimodal embeddings from the MIMIC-CXR dataset, we discover approximately 5,000 monosemantic patterns spanning cardiac, pulmonary, pleural, structural, device, and artifact categories. Each pattern exhibits consistent activation behavior across images sharing specific radiological features, enabling transparent attribution where predictions decompose into 20-50 interpretable patterns with verifiable activation galleries. CXR-LanIC achieves competitive diagnostic accuracy on five key findings while providing the foundation for natural language explanations through planned large multimodal model annotation. Our key innovation lies in extracting interpretable features from a classifier trained on specific diagnostic objectives rather than general-purpose embeddings, ensuring discovered patterns are directly relevant to clinical decision-making, demonstrating that medical AI systems can be both accurate and interpretable, supporting safer clinical deployment through transparent, clinically grounded explanations.

Reinforcement learning with verifiable rewards (RLVR) has become the mainstream technique for training LLM agents. However, RLVR highly depends on well-crafted task queries and corresponding ground-truth answers to provide accurate rewards, which requires significant human effort and hinders the scaling of RL processes, especially in agentic scenarios. Although a few recent works explore task synthesis methods, the difficulty of generated agentic tasks can hardly be controlled to provide effective RL training advantages. To achieve agentic RLVR with higher scalability, we explore self-play training for deep search agents, in which the learning LLM utilizes multi-turn search engine calling and acts simultaneously as both a task proposer and a problem solver. The task proposer aims to generate deep search queries with well-defined ground-truth answers and increasing task difficulty. The problem solver tries to handle the generated search queries and output the correct answer predictions. To ensure that each generated search query has accurate ground truth, we collect all the searching results from the proposer's trajectory as external knowledge, then conduct retrieval-augmentation generation (RAG) to test whether the proposed query can be correctly answered with all necessary search documents provided. In this search self-play (SSP) game, the proposer and the solver co-evolve their agent capabilities through both competition and cooperation. With substantial experimental results, we find that SSP can significantly improve search agents' performance uniformly on various benchmarks without any supervision under both from-scratch and continuous RL training setups. The code is at https://github.com/Qwen-Applications/SSP.

Archaeological predictive modelling estimates where undiscovered sites are likely to occur by combining known locations with environmental and geospatial variables, presenting a positive-unlabeled (PU) learning challenge where confirmed sites are rare and most locations are unlabeled rather than truly negative. To overcome this, we propose asymmetric dual pseudolabeling (DPL), an end-to-end deep learning method that learns from sparse positives directly from multi-band geospatial imagery without hand-crafted feature engineering or assumptions about site absence, and evaluate on two prominent archaeological datasets. On the Sagalassos dataset, evaluated against an independent, held-out field survey, DPL outperforms the LAMAP baseline by 12% in F1 and 29% in Recall, while LAMAP maintains advantages in probability ranking. Standard supervised baselines fail catastrophically when negatives are uncertain; positive-only training collapses to predicting everywhere, es- tablishing empirical bounds. On the Cyprus dataset, a pure PU setting without confirmed negatives, SL inverts probability rankings while DPL recovers discrimination. DPL ensembles produce interpretable probability surfaces supporting survey planning, enabling effective site discovery from minimal labeled data.

We present an experimental recipe for studying the relationship between training data and language model (LM) behavior. We outline steps for intervening on data batches -- i.e., ``rewriting history'' -- and then retraining model checkpoints over that data to test hypotheses relating data to behavior. Our recipe breaks down such an intervention into stages that include selecting evaluation items from a benchmark that measures model behavior, matching relevant documents to those items, and modifying those documents before retraining and measuring the effects. We demonstrate the utility of our recipe through case studies on factual knowledge acquisition in LMs, using both cooccurrence statistics and information retrieval methods to identify documents that might contribute to knowledge learning. Our results supplement past observational analyses that link cooccurrence to model behavior, while demonstrating that extant methods for identifying relevant training documents do not fully explain an LM's ability to correctly answer knowledge questions. Overall, we outline a recipe that researchers can follow to test further hypotheses about how training data affects model behavior. Our code is made publicly available to promote future work.

Generative AI systems are increasingly assisting and acting on behalf of end users in practical settings, from digital shopping assistants to next-generation autonomous cars. In this context, safety is no longer about blocking harmful content, but about preempting downstream hazards like financial or physical harm. Yet, most AI guardrails continue to rely on output classification based on labeled datasets and human-specified criteria,making them brittle to new hazardous situations. Even when unsafe conditions are flagged, this detection offers no path to recovery: typically, the AI system simply refuses to act--which is not always a safe choice. In this work, we argue that agentic AI safety is fundamentally a sequential decision problem: harmful outcomes arise from the AI system's continually evolving interactions and their downstream consequences on the world. We formalize this through the lens of safety-critical control theory, but within the AI model's latent representation of the world. This enables us to build predictive guardrails that (i) monitor an AI system's outputs (actions) in real time and (ii) proactively correct risky outputs to safe ones, all in a model-agnostic manner so the same guardrail can be wrapped around any AI model. We also offer a practical training recipe for computing such guardrails at scale via safety-critical reinforcement learning. Our experiments in simulated driving and e-commerce settings demonstrate that control-theoretic guardrails can reliably steer LLM agents clear of catastrophic outcomes (from collisions to bankruptcy) while preserving task performance, offering a principled dynamic alternative to today's flag-and-block guardrails.

Large Language Models (LLMs) process every token through all layers of a transformer stack, causing wasted computation on simple queries and insufficient flexibility for harder ones that need deeper reasoning. Adaptive-depth methods can improve efficiency, but prior approaches rely on costly inference-time search, architectural changes, or large-scale retraining, and in practice often degrade accuracy despite efficiency gains. We introduce Dr. LLM, Dynamic routing of Layers for LLMs, a retrofittable framework that equips pretrained models with lightweight per-layer routers deciding to skip, execute, or repeat a block. Routers are trained with explicit supervision: using Monte Carlo Tree Search (MCTS), we derive high-quality layer configurations that preserve or improve accuracy under a compute budget. Our design, windowed pooling for stable routing, focal loss with class balancing, and bottleneck MLP routers, ensures robustness under class imbalance and long sequences. On ARC (logic) and DART (math), Dr. LLM improves accuracy by up to +3.4%p while saving 5 layers per example on average. Routers generalize to out-of-domain tasks (MMLU, GSM8k, AIME, TruthfulQA, SQuADv2, GPQA, PIQA, AGIEval) with only 0.85% accuracy drop while retaining efficiency, and outperform prior routing methods by up to +7.7%p. Overall, Dr. LLM shows that explicitly supervised routers retrofit frozen LLMs for budget-aware, accuracy-driven inference without altering base weights. Code is available at https://github.com/parameterlab/dr-llm.

Spatial Transcriptomics (ST) enables the measurement of gene expression while preserving spatial information, offering critical insights into tissue architecture and disease pathology. Recent developments have explored the use of hematoxylin and eosin (H&E)-stained whole-slide images (WSIs) to predict transcriptome-wide gene expression profiles through deep neural networks. This task is commonly framed as a regression problem, where each input corresponds to a localized image patch extracted from the WSI. However, predicting spatial gene expression from histological images remains a challenging problem due to the significant modality gap between visual features and molecular signals. Recent studies have attempted to incorporate both local and global information into predictive models. Nevertheless, existing methods still suffer from two key limitations: (1) insufficient granularity in local feature extraction, and (2) inadequate coverage of global spatial context. In this work, we propose a novel framework, MMAP (Multi-MAgnification and Prototype-enhanced architecture), that addresses both challenges simultaneously. To enhance local feature granularity, MMAP leverages multi-magnification patch representations that capture fine-grained histological details. To improve global contextual understanding, it learns a set of latent prototype embeddings that serve as compact representations of slide-level information. Extensive experimental results demonstrate that MMAP consistently outperforms all existing state-of-the-art methods across multiple evaluation metrics, including Mean Absolute Error (MAE), Mean Squared Error (MSE), and Pearson Correlation Coefficient (PCC).

Training web agents to navigate complex, real-world websites requires them to master $\textit{subtasks}$ - short-horizon interactions on multiple UI components (e.g., choosing the correct date in a date picker, or scrolling in a container to extract information). We introduce WARC-Bench (Web Archive Benchmark), a novel web navigation benchmark featuring 438 tasks designed to evaluate multimodal AI agents on subtasks. WARC-Bench enables sandboxed interactions with dynamic and realistic webpages using Web ARChive files. We show that WARC-Bench is challenging for leading computer-use models, with the highest observed success rate being 64.8%. To improve open source models on subtask, we explore two common training techniques: supervised fine-tuning (SFT) and reinforcement learning with verifiable rewards (RLVR). Experiments show that SFT models obtain a 48.8% success rate on the benchmark. Training with RLVR over SFT checkpoints, even in data-scarce settings, improves the score to 52.8% on WARC-Bench, outperforming many frontier models. Our analysis concludes that mastering these subtasks is essential for robust web planning and navigation, and is a capability not extensively evaluated by existing benchmarks.

This paper takes a closer look at Git Re-Basin, an interesting new approach to merge trained models. We propose a hierarchical model merging scheme that significantly outperforms the standard MergeMany algorithm. With our new algorithm, we find that Re-Basin induces adversarial and perturbation robustness into the merged models, with the effect becoming stronger the more models participate in the hierarchical merging scheme. However, in our experiments Re-Basin induces a much bigger performance drop than reported by the original authors.

We introduce CAPC-CG, the Chinese Adaptive Policy Communication (Central Government) Corpus, the first open dataset of Chinese policy directives annotated with a five-color taxonomy of clear and ambiguous language categories, building on Ang's theory of adaptive policy communication. Spanning 1949-2023, this corpus includes national laws, administrative regulations, and ministerial rules issued by China's top authorities. Each document is segmented into paragraphs, producing a total of 3.3 million units. Alongside the corpus, we release comprehensive metadata, a two-round labeling framework, and a gold-standard annotation set developed by expert and trained coders. Inter-annotator agreement achieves a Fleiss's kappa of K = 0.86 on directive labels, indicating high reliability for supervised modeling. We provide baseline classification results with several large language models (LLMs), together with our annotation codebook, and describe patterns from the dataset. This release aims to support downstream tasks and multilingual NLP research in policy communication.

Effective removal of semantic watermarks requires balancing three competing objectives: \emph{high removal success}, \emph{low perceptual distortion}, and \emph{low computational cost}. However, existing single-image attacks typically optimize only for the first two, achieving strong watermark suppression but relying on expensive, multi-step optimization that limits practical deployment. In this work, we show that this trade-off is fundamental: no current approach achieves all three properties simultaneously. We introduce \textsc{DAWN}, a lightweight, training-free attack that explicitly targets the low-cost regime while maintaining competitive removal performance. \textsc{DAWN} works by projecting a watermarked image onto natural-image priors in complementary frequency and semantic spaces, suppressing watermark signals that deviate from natural statistics, and then applying a decoupled perceptual-alignment step to restore visual consistency with minimal artifact. Across diverse pixel-, frequency-, and latent-space watermarking schemes, \textsc{DAWN} consistently reduces detectability while preserving structural and semantic fidelity, demonstrating that efficient, low-resource watermark removal is feasible with only modest perceptual degradation. Our code is available at https://github.com/Pragati-Meshram/DAWN.

Large Language Model (LLM)-powered Multi-agent systems (MAS) have achieved state-of-the-art results on various complex reasoning tasks. Recent works have proposed techniques to automate the design of MASes, eliminating the need for manual engineering. However, these techniques perform poorly, often achieving similar or inferior performance to simple baselines. Furthermore, they require computationally expensive re-discovery of architectures for each new task domain and expensive data annotation on domains without existing labeled validation sets. A critical insight is that simple Chain of Thought (CoT) reasoning often performs competitively with these complex systems, suggesting that the fundamental reasoning unit of MASes, CoT, warrants further investigation. To this end, we present a new paradigm for automatic MAS design that pivots the focus to optimizing CoT reasoning. We introduce the Agentic Reasoning Module (ARM), an agentic generalization of CoT where each granular reasoning step is executed by a specialized reasoning module. This module is discovered through a tree search over the code space, starting from a simple CoT module and evolved using mutations informed by reflection on execution traces. The resulting ARM acts as a versatile reasoning building block which can be utilized as a direct recursive loop or as a subroutine in a learned meta-orchestrator. Our approach significantly outperforms both manually designed MASes and state-of-the-art automatic MAS design methods. Crucially, MASes built with ARM exhibit superb generalization, maintaining high performance across different foundation models and task domains without further optimization.

Organizing large-scale patent corpora according to classification schemes is a core information management task that determines the accuracy and efficiency of prior art retrieval, technology knowledge discovery, and intellectual property decision-making. Recent approaches distill natural language rationales generated by large language models (LLMs) into compact student models, yet logical errors, label mismatches, and taxonomy misalignments inherent in these rationales are indiscriminately absorbed during training, undermining classification reliability and propagating errors throughout downstream information processes. Rather than correcting such errors post-hoc, we propose Self-Filtered Distillation (SFD), which embeds quality assurance directly into the learning process by reinterpreting LLM-generated rationales as trust indicators rather than ground-truth supervision. SFD integrates three unsupervised signals into a unified trust score that dynamically modulates each training instance's contribution: Self-Consistency, which quantifies agreement among independently generated rationales; Class Entailment Alignment, which evaluates semantic coherence between a rationale and its assigned CPC class definition; and LLM Agreement Scoring, which assesses external plausibility through an independent verifier. On the USPTO-2M benchmark comprising over two million patents, SFD achieves up to 38.7\% relative improvement in Macro-F1 across four student architectures, and the strong correlation between trust scores and expert judgments ($r = 0.685$) confirms that the framework provides not only accurate predictions but also decomposable confidence semantics that enable auditable and self-documenting classification outcomes for large-scale patent knowledge organization.

The task of learning a diffusion-based neural sampler for drawing samples from an unnormalized target distribution can be viewed as a stochastic optimal control problem on path measures. However, the training of neural samplers can be challenging when the target distribution is multimodal with significant barriers separating the modes, potentially leading to mode collapse. We propose a framework named Proximal Diffusion Neural Sampler (PDNS) that addresses these challenges by tackling the stochastic optimal control problem via proximal point method on the space of path measures. PDNS decomposes the learning process into a series of simpler subproblems that create a path gradually approaching the desired distribution. This staged procedure traces a progressively refined path to the desired distribution and promotes thorough exploration across modes. For a practical and efficient realization, we instantiate each proximal step with a proximal weighted denoising cross-entropy (WDCE) objective. We demonstrate the effectiveness and robustness of PDNS through extensive experiments on both continuous and discrete sampling tasks, including challenging scenarios in molecular dynamics and statistical physics. Our code is available at https://github.com/AlexandreGUO2001/PDNS.

Autonomous web agents are increasingly deployed for long-horizon tasks, yet their ability to adhere to real-world policies remains critically underexplored compared to standard safety objectives. To address this gap, we introduce PolicyGuardBench, a benchmark of 60k policy-trajectory pairs designed to evaluate compliance through both full-trajectory and novel prefix-based violation detection tasks. Using this dataset, we train PolicyGuard, a lightweight guardrail model that achieves strong detection accuracy while maintaining high inference efficiency. Notably, our model demonstrates robust generalization capabilities, preserving high performance even on unseen domains. These contributions establish a comprehensive framework for studying policy compliance, showing that accurate and generalizable guardrails are feasible at small scales.

With the rapid progress of multi-agent large language model (LLM) reasoning, how to effectively aggregate answers from multiple LLMs has emerged as a fundamental challenge. Standard majority voting treats all answers equally, failing to consider latent heterogeneity and correlation across models. In this work, we design two new aggregation algorithms called Optimal Weight (OW) and Inverse Surprising Popularity (ISP), leveraging both first-order and second-order information. Our theoretical analysis shows these methods provably mitigate inherent limitations of majority voting under mild assumptions, leading to more reliable collective decisions. We empirically validate our algorithms on synthetic datasets, popular LLM fine-tuning benchmarks such as UltraFeedback and MMLU, and a real-world healthcare setting ARMMAN. Our algorithms consistently outperform standard baselines, establishing a robust, training-free framework for effective multi-agent LLM aggregation.

High-sensitivity clutter filtering is a fundamental step in ultrasound microvascular imaging. Singular value decomposition (SVD) and robust principal component analysis (rPCA) are the main clutter filtering strategies. However, both strategies are limited in feature modeling and separation of tissue and blood flow for high-quality microvascular imaging. Recently, deep learning-based clutter filtering has shown potential in more thoroughly separating tissue and blood flow signals. However, the existing supervised filters face the lack of interpretability and the training ground truth. While the interpretability issue can be addressed by algorithm deep unfolding, the training ground truth remains unsolved. This paper proposes an unsupervised unfolded rPCA (U2-rPCA) method that preserves mathematical interpretability and is insusceptible to learning labels. Specifically, U2-rPCA is unfolded from an iteratively reweighted least squares (IRLS) rPCA baseline with intrinsic low-rank and sparse regularization. In addition, a sparse-enhancement unit is plugged into the network to strengthen its capability to capture the sparse micro-flow signals. U2-rPCA is like an adaptive filter that is trained with part of the image sequence and then used for the following frames. Experimental validations on a in-silico dataset and public in-vivo datasets demonstrated the outperformance of U2-rPCA when compared with the SVD filter, the rPCA baseline, and another deep learning-based filter. Particularly, the proposed method improved the contrast-to-noise ratio (CNR) of the power Doppler image by 1.91 dB to 8.48 dB compared to other methods. Furthermore, the effectiveness of the building modules of U2-rPCA was validated through ablation studies.

Using Large Language Models to produce intermediate thoughts, a.k.a. Chain-of-thought (CoT), before providing an answer has been a successful recipe for solving complex language tasks. In robotics, similar embodied CoT strategies, generating thoughts before actions, have also been shown to lead to improved performance when using Vision-Language-Action models (VLAs). As these techniques increase the length of the model's generated outputs to include the thoughts, the inference time is negatively affected. Delaying an agent's actions in real-world executions, as in robotic manipulation settings, strongly affects the usability of a method, as tasks require long sequences of actions. However, is the generation of long chains-of-thought a strong prerequisite for achieving performance improvements? In this work, we explore the idea of Hybrid Training (HyT), a framework that enables VLAs to learn from thoughts and benefit from the associated performance gains, while enabling the possibility to leave out CoT generation during inference. Furthermore, by learning to conditionally predict a diverse set of outputs, HyT supports flexibility at inference time, enabling the model to either predict actions directly, generate thoughts or follow instructions. We evaluate the proposed method in a series of simulated benchmarks and real-world experiments.

Can visual imagery be driven solely by language? This idea goes against cognitive science's traditional view that visual mental imagery is only possible through pictorial representations. Large Language Models (LLMs) provide nascent evidence not only that visual mental imagery via propositional-representations is possible, but that it can be more robust than human imagination. We created dozens of novel items for an extension to a classic task which is argued to be solvable exclusively via pictorial representations (i.e., language alone would be insufficient). Subjects were asked to imagine a series of compositional letter and shape transformations and identify the resultant "image". We found that the best LLMs performed significantly better than humans ($n = 100$ human participants, $p < .0001$), indicating the existence of an artificial phantasia, or emergent "visual" mental imagery that may not be pictorial. Furthermore, we tested reasoning models with variable reasoning-token allocation and found that models perform best with longer reasoning chains, demonstrating a linguistic impact on the task -- language alone may be sufficient. We examined three emergent imagery hypotheses: pure propositional imagery, propositional imagery with visio-linguistic priors, or pictorial visual imagery (classical visual imagery). Our study not only presents evidence for a previously unreported emergent cognitive capacity of LLMs, but also reignites debate on the requirement for a pictorial format in mental imagery.

Along with the rapid advancement of numerous Text-to-Video (T2V) models, growing concerns have emerged regarding their safety risks. While recent studies have explored vulnerabilities in models like LLMs, VLMs, and Text-to-Image (T2I) models through jailbreak attacks, T2V models remain largely unexplored, leaving a significant safety gap. To address this gap, we introduce SceneSplit, a novel black-box jailbreak method that works by fragmenting a harmful narrative into multiple scenes, each individually benign. This approach manipulates the generative output space, the abstract set of all potential video outputs for a given prompt, using the combination of scenes as a powerful constraint to guide the final outcome. While each scene individually corresponds to a wide and safe space where most outcomes are benign, their sequential combination collectively restricts this space, narrowing it to an unsafe region and significantly increasing the likelihood of generating a harmful video. This core mechanism is further enhanced through iterative scene manipulation, which bypasses the safety filter within this constrained unsafe region. Additionally, a strategy library that reuses successful attack patterns further improves the attack's overall effectiveness and robustness. To validate our method, we evaluate SceneSplit across 11 safety categories from T2VSafetyBench on T2V models. Our results show that it achieves a high average Attack Success Rate (ASR) of 77.2% on Luma Ray2, 84.1% on Hailuo, 78.2% on Veo2, 78.6% on Kling V1.0, and 68.6% on Sora2, significantly outperforming the existing baselines. Through this work, we demonstrate that current T2V safety mechanisms are vulnerable to attacks that exploit narrative structure, providing new insights for understanding and improving the safety of T2V models.

Recent advances in vision-language models (VLMs) have achieved remarkable performance on standard medical benchmarks, yet their true clinical reasoning ability remains unclear. Existing datasets predominantly emphasize classification accuracy, creating an evaluation illusion in which models appear proficient while still failing at high-stakes diagnostic reasoning. We introduce Neural-MedBench, a compact yet reasoning-intensive benchmark specifically designed to probe the limits of multimodal clinical reasoning in neurology. Neural-MedBench integrates multi-sequence MRI scans, structured electronic health records, and clinical notes, and encompasses three core task families: differential diagnosis, lesion recognition, and rationale generation. To ensure reliable evaluation, we develop a hybrid scoring pipeline that combines LLM-based graders, clinician validation, and semantic similarity metrics. Through systematic evaluation of state-of-the-art VLMs, including GPT-4o, Claude-4, and MedGemma, we observe a sharp performance drop compared to conventional datasets. Error analysis shows that reasoning failures, rather than perceptual errors, dominate model shortcomings. Our findings highlight the necessity of a Two-Axis Evaluation Framework: breadth-oriented large datasets for statistical generalization, and depth-oriented, compact benchmarks such as Neural-MedBench for reasoning fidelity. We release Neural-MedBench at https://neuromedbench.github.io/ as an open and extensible diagnostic testbed, which guides the expansion of future benchmarks and enables rigorous yet cost-effective assessment of clinically trustworthy AI.

Risk categorization in 10-K risk disclosures matters for oversight and investment, yet no public benchmark evaluates unsupervised topic models for this task. We present GRAB, a finance-specific benchmark with 1.61M sentences from 8,247 filings and span-grounded sentence labels produced without manual annotation by combining FinBERT token attention, YAKE keyphrase signals, and taxonomy-aware collocation matching. Labels are anchored in a risk taxonomy mapping 193 terms to 21 fine-grained types nested under five macro classes; the 21 types guide weak supervision, while evaluation is reported at the macro level. GRAB unifies evaluation with fixed dataset splits and robust metrics--Accuracy, Macro-F1, Topic BERTScore, and the entropy-based Effective Number of Topics. The dataset, labels, and code enable reproducible, standardized comparison across classical, embedding-based, neural, and hybrid topic models on financial disclosures.

Accurately forecasting the long-term evolution of turbulence represents a grand challenge in scientific computing and is crucial for applications ranging from climate modeling to aerospace engineering. Existing deep learning methods, particularly neural operators, often fail in long-term autoregressive predictions, suffering from catastrophic error accumulation and a loss of physical fidelity. This failure stems from their inability to simultaneously capture the distinct mathematical structures that govern turbulent dynamics: local, dissipative effects and global, non-local interactions. In this paper, we propose the {\textbf{\underline{D}}}ifferential-{\textbf{\underline{I}}}ntegral {\textbf{\underline{N}}}eural {\textbf{\underline{O}}}perator (\method{}), a novel framework designed from a first-principles approach of operator decomposition. \method{} explicitly models the turbulent evolution through parallel branches that learn distinct physical operators: a local differential operator, realized by a constrained convolutional network that provably converges to a derivative, and a global integral operator, captured by a Transformer architecture that learns a data-driven global kernel. This physics-based decomposition endows \method{} with exceptional stability and robustness. Through extensive experiments on the challenging 2D Kolmogorov flow benchmark, we demonstrate that \method{} significantly outperforms state-of-the-art models in long-term forecasting. It successfully suppresses error accumulation over hundreds of timesteps, maintains high fidelity in both the vorticity fields and energy spectra, and establishes a new benchmark for physically consistent, long-range turbulence forecast.