High-quality scientific illustrations are crucial for effectively communicating complex scientific and technical concepts, yet their manual creation remains a well-recognized bottleneck in both academia and industry. We present FigureBench, the first large-scale benchmark for generating scientific illustrations from long-form scientific texts. It contains 3,300 high-quality scientific text-figure pairs, covering diverse text-to-illustration tasks from scientific papers, surveys, blogs, and textbooks. Moreover, we propose AutoFigure, the first agentic framework that automatically generates high-quality scientific illustrations based on long-form scientific text. Specifically, before rendering the final result, AutoFigure engages in extensive thinking, recombination, and validation to produce a layout that is both structurally sound and aesthetically refined, outputting a scientific illustration that achieves both structural completeness and aesthetic appeal. Leveraging the high-quality data from FigureBench, we conduct extensive experiments to test the performance of AutoFigure against various baseline methods. The results demonstrate that AutoFigure consistently surpasses all baseline methods, producing publication-ready scientific illustrations. The code, dataset and huggingface space are released in https://github.com/ResearAI/AutoFigure.

Reliable localization is crucial for navigation in forests, where GPS is often degraded and LiDAR measurements are repetitive, occluded, and structurally complex. These conditions weaken the assumptions of traditional urban-centric localization methods, which assume that consistent features arise from unique structural patterns, necessitating forest-centric solutions to achieve robustness in these environments. To address these challenges, we propose TreeLoc, a LiDAR-based global localization framework for forests that handles place recognition and 6-DoF pose estimation. We represent scenes using tree stems and their Diameter at Breast Height (DBH), which are aligned to a common reference frame via their axes and summarized using the tree distribution histogram (TDH) for coarse matching, followed by fine matching with a 2D triangle descriptor. Finally, pose estimation is achieved through a two-step geometric verification. On diverse forest benchmarks, TreeLoc outperforms baselines, achieving precise localization. Ablation studies validate the contribution of each component. We also propose applications for long-term forest management using descriptors from a compact global tree database. TreeLoc is open-sourced for the robotics community at https://github.com/minwoo0611/TreeLoc.

How close are neural networks to the best they could possibly do? Standard benchmarks cannot answer this because they lack access to the true posterior p(y|x). We use class-conditional normalizing flows as oracles that make exact posteriors tractable on realistic images (AFHQ, ImageNet). This enables five lines of investigation. Scaling laws: Prediction error decomposes into irreducible aleatoric uncertainty and reducible epistemic error; the epistemic component follows a power law in dataset size, continuing to shrink even when total loss plateaus. Limits of learning: The aleatoric floor is exactly measurable, and architectures differ markedly in how they approach it: ResNets exhibit clean power-law scaling while Vision Transformers stall in low-data regimes. Soft labels: Oracle posteriors contain learnable structure beyond class labels: training with exact posteriors outperforms hard labels and yields near-perfect calibration. Distribution shift: The oracle computes exact KL divergence of controlled perturbations, revealing that shift type matters more than shift magnitude: class imbalance barely affects accuracy at divergence values where input noise causes catastrophic degradation. Active learning: Exact epistemic uncertainty distinguishes genuinely informative samples from inherently ambiguous ones, improving sample efficiency. Our framework reveals that standard metrics hide ongoing learning, mask architectural differences, and cannot diagnose the nature of distribution shift.

Neural Radiance Fields (NeRFs) have emerged as a powerful paradigm for multi-view reconstruction, complementing classical photogrammetric pipelines based on Structure-from-Motion (SfM) and Multi-View Stereo (MVS). However, their reliability for quantitative 3D analysis in dense, self-occluding scenes remains poorly understood. In this study, we identify a fundamental failure mode of implicit density fields under heavy occlusion, which we term Interior Geometric Degradation (IGD). We show that transmittance-based volumetric optimization satisfies photometric supervision by reconstructing hollow or fragmented structures rather than solid interiors, leading to systematic instance undercounting. Through controlled experiments on synthetic datasets with increasing occlusion, we demonstrate that state-of-the-art mask-supervised NeRFs saturate at approximately 89% instance recovery in dense scenes, despite improved surface coherence and mask quality. To overcome this limitation, we introduce an explicit geometric pipeline based on Sparse Voxel Rasterization (SVRaster), initialized from SfM feature geometry. By projecting 2D instance masks onto an explicit voxel grid and enforcing geometric separation via recursive splitting, our approach preserves physical solidity and achieves a 95.8% recovery rate in dense clusters. A sensitivity analysis using degraded segmentation masks further shows that explicit SfM-based geometry is substantially more robust to supervision failure, recovering 43% more instances than implicit baselines. These results demonstrate that explicit geometric priors are a prerequisite for reliable quantitative analysis in highly self-occluding 3D scenes.

Social infrastructure and other built environments are increasingly expected to support well-being and community resilience by enabling social interaction. Yet in civil and built-environment research, there is no consistent and privacy-preserving way to represent and measure socially meaningful interaction in these spaces, leaving studies to operationalize "interaction" differently across contexts and limiting practitioners' ability to evaluate whether design interventions are changing the forms of interaction that social capital theory predicts should matter. To address this field-level and methodological gap, we introduce the Dyadic User Engagement DataseT (DUET) dataset and an embedded kinesics recognition framework that operationalize Ekman and Friesen's kinesics taxonomy as a function-level interaction vocabulary aligned with social capital-relevant behaviors (e.g., reciprocity and attention coordination). DUET captures 12 dyadic interactions spanning all five kinesic functions-emblems, illustrators, affect displays, adaptors, and regulators-across four sensing modalities and three built-environment contexts, enabling privacy-preserving analysis of communicative intent through movement. Benchmarking six open-source, state-of-the-art human activity recognition models quantifies the difficulty of communicative-function recognition on DUET and highlights the limitations of ubiquitous monadic, action-level recognition when extended to dyadic, socially grounded interaction measurement. Building on DUET, our recognition framework infers communicative function directly from privacy-preserving skeletal motion without handcrafted action-to-function dictionaries; using a transfer-learning architecture, it reveals structured clustering of kinesic functions and a strong association between representation quality and classification performance while generalizing across subjects and contexts.

Robot-assisted rehabilitation offers an effective approach, wherein exoskeletons adapt to users' needs and provide personalized assistance. However, to deliver such assistance, accurate prediction of the user's joint torques is essential. In this work, we propose a feature extraction pipeline using 8-channel surface electromyography (sEMG) signals to predict elbow and shoulder joint torques. For preliminary evaluation, this pipeline was integrated into two neural network models: the Multilayer Perceptron (MLP) and the Temporal Convolutional Network (TCN). Data were collected from a single subject performing elbow and shoulder movements under three load conditions (0 kg, 1.10 kg, and 1.85 kg) using three motion-capture cameras. Reference torques were estimated from center-of-mass kinematics under the assumption of static equilibrium. Our offline analyses showed that, with our feature extraction pipeline, MLP model achieved mean RMSE of 0.963 N m, 1.403 N m, and 1.434 N m (over five seeds) for elbow, front-shoulder, and side-shoulder joints, respectively, which were comparable to the TCN performance. These results demonstrate that the proposed feature extraction pipeline enables a simple MLP to achieve performance comparable to that of a network designed explicitly for temporal dependencies. This finding is particularly relevant for applications with limited training data, a common scenario patient care.

Self-evolution methods enhance code generation through iterative "generate-verify-refine" cycles, yet existing approaches suffer from low exploration efficiency, failing to discover solutions with superior complexity within limited budgets. This inefficiency stems from initialization bias trapping evolution in poor solution regions, uncontrolled stochastic operations lacking feedback guidance, and insufficient experience utilization across tasks. To address these bottlenecks, we propose Controlled Self-Evolution (CSE), which consists of three key components. Diversified Planning Initialization generates structurally distinct algorithmic strategies for broad solution space coverage. Genetic Evolution replaces stochastic operations with feedback-guided mechanisms, enabling targeted mutation and compositional crossover. Hierarchical Evolution Memory captures both successful and failed experiences at inter-task and intra-task levels. Experiments on EffiBench-X demonstrate that CSE consistently outperforms all baselines across various LLM backbones. Furthermore, CSE achieves higher efficiency from early generations and maintains continuous improvement throughout evolution. Our code is publicly available at https://github.com/QuantaAlpha/EvoControl.

Although recent years have seen significant progress of humanoid robots in walking and running, the frequent foot strikes with ground during these locomotion gaits inevitably generate high instantaneous impact forces, which leads to exacerbated joint wear and poor energy utilization. Roller skating, as a sport with substantial biomechanical value, can achieve fast and continuous sliding through rational utilization of body inertia, featuring minimal kinetic energy loss. Therefore, this study proposes a novel humanoid robot with each foot equipped with a row of four passive wheels for roller skating. A deep reinforcement learning control framework is also developed for the swizzle gait with the reward function design based on the intrinsic characteristics of roller skating. The learned policy is first analyzed in simulation and then deployed on the physical robot to demonstrate the smoothness and efficiency of the swizzle gait over traditional bipedal walking gait in terms of Impact Intensity and Cost of Transport during locomotion. A reduction of $75.86\%$ and $63.34\%$ of these two metrics indicate roller skating as a superior locomotion mode for enhanced energy efficiency and joint longevity.

We address the problem of classifying trajectory data generated by some nonlinear dynamics, where each class corresponds to a distinct dynamical system. We propose Dynafit, a kernel-based method for learning a distance metric between training trajectories and the underlying dynamics. New observations are assigned to the class with the most similar dynamics according to the learned metric. The learning algorithm approximates the Koopman operator which globally linearizes the dynamics in a (potentially infinite) feature space associated with a kernel function. The distance metric is computed in feature space independently of its dimensionality by using the kernel trick common in machine learning. We also show that the kernel function can be tailored to incorporate partial knowledge of the dynamics when available. Dynafit is applicable to various classification tasks involving nonlinear dynamical systems and sensors. We illustrate its effectiveness on three examples: chaos detection with the logistic map, recognition of handwritten dynamics and of visual dynamic textures.

Large language models (LLMs) excel at natural language tasks but remain brittle in domains requiring precise logical and symbolic reasoning. Chaotic dynamical systems provide an especially demanding test because chaos is deterministic yet often misinterpreted as randomness or complexity. We introduce ChaosBench-Logic, a benchmark that evaluates LLM reasoning across 30 diverse dynamical systems using a unified first-order logic (FOL) ontology. Each system is annotated with truth assignments for 11 semantic predicates, and 621 questions are generated across seven reasoning categories, including multi-hop implications, cross-system analogies, counterfactual reasoning, bias probes, and multi-turn dialogues. We define metrics for logical accuracy, implication consistency, dialogue coherence, and contradiction, and we release an open-source evaluation pipeline. Initial experiments show that frontier LLMs such as GPT-4, Claude 3.5 Sonnet, Gemini 2.5 Flash, and the open-source LLaMA-3 70B achieve 91-94% per-item accuracy, yet still score 0% on compositional items and exhibit fragile global coherence. Dialogue-level accuracy ranges from 53.1% (GPT-4 CoT) to 75.5% (LLaMA-3 zero-shot). ChaosBench-Logic provides a rigorous testbed for diagnosing such failures and a foundation for developing neuro-symbolic approaches that improve scientific reasoning in LLMs.

We present Racka, a lightweight, continually pretrained large language model designed to bridge the resource gap between Hungarian and high-resource languages such as English and German. Racka employs parameter-efficient continual pretraining via Low-Rank Adaptation (LoRA) on a Qwen-3 4B backbone, making the recipe practical on A100 (40GB)-based HPC clusters with low inter-node bandwidth. To better match the training distribution, we replace and adapt the tokenizer, achieving substantially improved tokenization fertility for Hungarian while maintaining competitive performance in English and German. The model is trained on 160B subword tokens drawn from a mixture of internet and high-quality curated sources, with a composition of 44% Hungarian, 24% English, 21% German, and 11% code. This data mix is chosen to mitigate catastrophic forgetting and preserve high-resource language capabilities during continual pretraining. Our preliminary results indicate modest but stable results in language adaptation.

Recent advances in hierarchical robot systems leverage a high-level planner to propose task plans and a low-level policy to generate robot actions. This design allows training the planner on action-free or even non-robot data sources (e.g., videos), providing transferable high-level guidance. Nevertheless, grounding these high-level plans into executable actions remains challenging, especially with the limited availability of high-quality robot data. To this end, we propose to improve the low-level policy through online interactions. Specifically, our approach collects online rollouts, retrospectively annotates the corresponding high-level goals from achieved outcomes, and aggregates these hindsight-relabeled experiences to update a goal-conditioned imitation policy. Our method, Hindsight Flow-conditioned Online Imitation (HinFlow), instantiates this idea with 2D point flows as the high-level planner. Across diverse manipulation tasks in both simulation and physical world, our method achieves more than $2\times$ performance improvement over the base policy, significantly outperforming the existing methods. Moreover, our framework enables policy acquisition from planners trained on cross-embodiment video data, demonstrating its potential for scalable and transferable robot learning.

Vision foundation models have demonstrated strong generalization in medical image segmentation by leveraging large-scale, heterogeneous pretraining. However, they often struggle to generalize to specialized clinical tasks under limited annotations or rare pathological variations, due to a mismatch between general priors and task-specific requirements. To address this, we propose Uncertainty-informed Collaborative Learning (UnCoL), a dual-teacher framework that harmonizes generalization and specialization in semi-supervised medical image segmentation. Specifically, UnCoL distills both visual and semantic representations from a frozen foundation model to transfer general knowledge, while concurrently maintaining a progressively adapting teacher to capture fine-grained and task-specific representations. To balance guidance from both teachers, pseudo-label learning in UnCoL is adaptively regulated by predictive uncertainty, which selectively suppresses unreliable supervision and stabilizes learning in ambiguous regions. Experiments on diverse 2D and 3D segmentation benchmarks show that UnCoL consistently outperforms state-of-the-art semi-supervised methods and foundation model baselines. Moreover, our model delivers near fully supervised performance with markedly reduced annotation requirements.

Gastrointestinal (GI) imaging via Wireless Capsule Endoscopy (WCE) generates a large number of images requiring manual screening. Deep learning-based Clinical Decision Support (CDS) systems can assist screening, yet their performance relies on the existence of large, diverse, training medical datasets. However, the scarcity of such data, due to privacy constraints and annotation costs, hinders CDS development. Generative machine learning offers a viable solution to combat this limitation. While current Synthetic Data Generation (SDG) methods, such as Generative Adversarial Networks and Variational Autoencoders have been explored, they often face challenges with training stability and capturing sufficient visual diversity, especially when synthesizing abnormal findings. This work introduces a novel VAE-based methodology for medical image synthesis and presents its application for the generation of WCE images. The novel contributions of this work include a) multiscale extension of the Vector Quantized VAE model, named as Multiscale Vector Quantized Variational Autoencoder (MSVQ-VAE); b) unlike other VAE-based SDG models for WCE image generation, MSVQ-VAE is used to seamlessly introduce abnormalities into normal WCE images; c) it enables conditional generation of synthetic images, enabling the introduction of different types of abnormalities into the normal WCE images; d) it performs experiments with a variety of abnormality types, including polyps, vascular and inflammatory conditions. The utility of the generated images for CDS is assessed via image classification. Comparative experiments demonstrate that training a CDS classifier using the abnormal images generated by the proposed methodology yield comparable results with a classifier trained with only real data. The generality of the proposed methodology promises its applicability to various domains related to medical multimedia.

While Chain-of-Thought (CoT) prompting enables sophisticated symbolic reasoning in LLMs, it remains confined to discrete text and cannot simulate the continuous, physics-governed dynamics of the real world. Recent video generation models have emerged as potential world simulators through Chain-of-Frames (CoF) reasoning -- materializing thought as frame-by-frame visual sequences, with each frame representing a physically-grounded reasoning step. Despite compelling demonstrations, a challenge persists: existing benchmarks, focusing on fidelity or alignment, do not assess CoF reasoning and thus cannot measure core cognitive abilities in multi-step planning, algorithmic logic, or abstract pattern extrapolation. This evaluation void prevents systematic understanding of model capabilities and principled guidance for improvement. We introduce Gen-ViRe (Generative Visual Reasoning Benchmark), a framework grounded in cognitive science and real-world AI applications, which decomposes CoF reasoning into six cognitive dimensions -- from perceptual logic to abstract planning -- and 24 subtasks. Through multi-source data curation, minimal prompting protocols, and hybrid VLM-assisted evaluation with detailed criteria, Gen-ViRe delivers the first quantitative assessment of video models as reasoners. Our experiments on SOTA systems reveal substantial discrepancies between impressive visual quality and actual reasoning depth, establishing baselines and diagnostic tools to advance genuine world simulators.

This position paper envisions a next-generation elderly monitoring system that moves beyond fall detection toward the broader goal of Activities of Daily Living (ADL) recognition. Our ultimate aim is to design privacy-preserving, edge-deployed, and federated AI systems that can robustly detect and understand daily routines, supporting independence and dignity in aging societies. At present, ADL-specific datasets are still under collection. As a preliminary step, we demonstrate feasibility through experiments using the SISFall dataset and its GAN-augmented variants, treating fall detection as a proxy task. We report initial results on federated learning with non-IID conditions, and embedded deployment on Jetson Orin Nano devices. We then outline open challenges such as domain shift, data scarcity, and privacy risks, and propose directions toward full ADL monitoring in smart-room environments. This work highlights the transition from single-task detection to comprehensive daily activity recognition, providing both early evidence and a roadmap for sustainable and human-centered elderly care AI.

Large language models (LLMs) have achieved impressive results in high-resource languages like English, yet their effectiveness in low-resource and morphologically rich languages remains underexplored. In this paper, we present a comprehensive evaluation of seven cutting-edge LLMs -- including GPT-4o, GPT-4, Claude~3.5~Sonnet, LLaMA~3.1, Mistral~Large~2, LLaMA-2~Chat~13B, and Mistral~7B~Instruct -- on a new cross-lingual benchmark covering \textbf{Cantonese, Japanese, and Turkish}. Our benchmark spans four diverse tasks: open-domain question answering, document summarization, English-to-X translation, and culturally grounded dialogue. We combine \textbf{human evaluations} (rating fluency, factual accuracy, and cultural appropriateness) with automated metrics (e.g., BLEU, ROUGE) to assess model performance. Our results reveal that while the largest proprietary models (GPT-4o, GPT-4, Claude~3.5) generally lead across languages and tasks, significant gaps persist in culturally nuanced understanding and morphological generalization. Notably, GPT-4o demonstrates robust multilingual performance even on cross-lingual tasks, and Claude~3.5~Sonnet achieves competitive accuracy on knowledge and reasoning benchmarks. However, all models struggle to some extent with the unique linguistic challenges of each language, such as Turkish agglutinative morphology and Cantonese colloquialisms. Smaller open-source models (LLaMA-2~13B, Mistral~7B) lag substantially in fluency and accuracy, highlighting the resource disparity. We provide detailed quantitative results, qualitative error analysis, and discuss implications for developing more culturally aware and linguistically generalizable LLMs. Our benchmark and evaluation data are released to foster reproducibility and further research.

Industrial Anomaly Detection (IAD) is a cornerstone for ensuring operational safety, maintaining product quality, and optimizing manufacturing efficiency. However, the advancement of IAD algorithms is severely hindered by the limitations of existing public benchmarks. Current datasets often suffer from restricted category diversity and insufficient scale, leading to performance saturation and poor model transferability in complex, real-world scenarios. To bridge this gap, we introduce Real-IAD Variety, the largest and most diverse IAD benchmark. It comprises 198,950 high-resolution images across 160 distinct object categories. The dataset ensures unprecedented diversity by covering 28 industries, 24 material types, 22 color variations, and 27 defect types. Our extensive experimental analysis highlights the substantial challenges posed by this benchmark: state-of-the-art multi-class unsupervised anomaly detection methods suffer significant performance degradation (ranging from 10% to 20%) when scaled from 30 to 160 categories. Conversely, we demonstrate that zero-shot and few-shot IAD models exhibit remarkable robustness to category scale-up, maintaining consistent performance and significantly enhancing generalization across diverse industrial contexts. This unprecedented scale positions Real-IAD Variety as an essential resource for training and evaluating next-generation foundation IAD models.

Diffusion models achieve remarkable fidelity in image synthesis, yet precise control over their outputs for targeted editing remains challenging. A key step toward controllability is to identify interpretable directions in the model's latent representations that correspond to semantic attributes. Existing approaches for finding interpretable directions typically rely on sampling large sets of images or training auxiliary networks, which limits efficiency. We propose an analytical method that derives semantic editing directions directly from the pretrained parameters of diffusion models, requiring neither additional data nor fine-tuning. Our insight is that self-attention weight matrices encode rich structural information about the data distribution learned during training. By computing the eigenvectors of these weight matrices, we obtain robust and interpretable editing directions. Experiments demonstrate that our method produces high-quality edits across multiple datasets while reducing editing time significantly by 60% over current benchmarks.

Despite recent advances in Unmanned Underwater Vehicle (UUV) attitude control, existing methods still struggle with generalizability, robustness to real-world disturbances, and efficient deployment. To address the above challenges, this paper presents EasyUUV, a Large Language Model (LLM)-enhanced, universal, and lightweight simulation-to-reality reinforcement learning (RL) framework for robust attitude control of UUVs. EasyUUV combines parallelized RL training with a hybrid control architecture, where a learned policy outputs high-level attitude corrections executed by an adaptive S-Surface controller. A multimodal LLM is further integrated to adaptively tune controller parameters at runtime using visual and textual feedback, enabling training-free adaptation to unmodeled dynamics. Also, we have developed a low-cost 6-DoF UUV platform and applied an RL policy trained through efficient parallelized simulation. Extensive simulation and real-world experiments validate the effectiveness and outstanding performance of EasyUUV in achieving robust and adaptive UUV attitude control across diverse underwater conditions. To facilitate reproducibility and further research, the source code, LLM prompts, and supplementary video are provided in the following repositories: Homepage: https://360zmem.github.io/easyuuv/ Video:https://youtu.be/m2yLQzxiIL

Prototypical self-supervised learning methods consistently suffer from partial prototype collapse, where multiple prototypes converge to nearly identical representations. This undermines their central purpose -- providing diverse and informative targets to guide encoders toward rich representations -- and has led practitioners to over-parameterize prototype sets or add ad-hoc regularizers, which mitigate symptoms rather than address the root cause. We empirically trace the collapse to the joint optimization of encoders and prototypes, which encourages a type of shortcut learning: early in training prototypes drift toward redundant representations that minimize loss without necessarily enhancing representation diversity. To break the joint optimization, we introduce a fully decoupled training strategy that learns prototypes and encoders under separate objectives. Concretely, we model prototypes as a Gaussian mixture updated with an online EM-style procedure, independent of the encoder's loss. This simple yet principled decoupling eliminates prototype collapse without explicit regularization and yields consistently diverse prototypes and stronger downstream performance.

What does it truly mean for a language model to "reason"? Current evaluations reward models' correct standalone answers-but correctness alone reveals little about the process that produced them. We argue that reasoning should be understood not as a static chain of steps but as a dynamic trajectory in which ideas interact, clash, and evolve into integrated insights. Building on the philosophical tradition of dialectics, we introduce SIEV, a structured evaluation framework that assesses reasoning through explicit thesis-antithesis-synthesis interactions. SIEV produces interpretable trajectories that highlight key properties of reasoning-robustness to challenge, adaptability under conflict, and synthesis across competing viewpoints-dimensions that conventional correctness-based metrics cannot capture. Empirical results on GSM and MMLU demonstrate substantial gaps in the reasoning abilities of state-of-the-art models: for example, GPT-5-chat loses more than 40 points (out of 100) on GSM when evaluated through SIEV's process-oriented lens. By shifting focus from what answer a model gives to how it arrives there, SIEV enables a more transparent and principled distinction between structured reasoning and surface-level pattern generation offering a clearer foundation for assessing and understanding the reasoning capabilities of LLMs.

Developing Large Language Models (LLMs) to cooperate and compete effectively within multi-agent systems (MASs) is a critical step towards more advanced intelligence. While reinforcement learning (RL) has proven effective for enhancing reasoning in single-agent tasks, its extension to multi-turn, multi-agent scenarios remains underexplored due to the challenges of long-horizon credit assignment and agent-specific advantage estimation. To address these challenges, we introduce MARSHAL, an end-to-end RL framework that incentivizes Multi-Agent Reasoning through Self-play witH strAtegic LLMs in both cooperative and competitive games. MARSHAL features a turn-level advantage estimator that aligns learning signals with each interaction for credit assignment, and an agent-specific advantage normalization to stabilize multi-agent training. By learning with self-play across cooperative and competitive games, MARSHAL agents trained from Qwen3-4B develop strong strategic abilities, with up to 28.7% performance improvements in held-out games. More importantly, the capability acquired through self-play generalizes beyond games, yielding consistent performance gains of MASs in reasoning benchmarks. When integrated into leading MASs, our MARSHAL agent achieves significant zero-shot performance gains of up to 10.0% on AIME, 7.6% on GPQA-Diamond, and 3.5% on average across all benchmarks. These results establish self-play in strategic games as a powerful approach for developing generalizable multi-agent reasoning capabilities in LLMs.

Multi-instance image generation (MIG) remains a significant challenge for modern diffusion models due to key limitations in achieving precise control over object layout and preserving the identity of multiple distinct subjects. To address these limitations, we introduce ContextGen, a novel Diffusion Transformer framework for multi-instance generation that is guided by both layout and reference images. Our approach integrates two key technical contributions: a Contextual Layout Anchoring (CLA) mechanism that incorporates the composite layout image into the generation context to robustly anchor the objects in their desired positions, and Identity Consistency Attention (ICA), an innovative attention mechanism that leverages contextual reference images to ensure the identity consistency of multiple instances. To address the absence of a large-scale, high-quality dataset for this task, we introduce IMIG-100K, the first dataset to provide detailed layout and identity annotations specifically designed for Multi-Instance Generation. Extensive experiments demonstrate that ContextGen sets a new state-of-the-art, outperforming existing methods especially in layout control and identity fidelity.

Lack of large, well-annotated emotional speech corpora continues to limit the performance and robustness of speech emotion recognition (SER), particularly as models grow more complex and the demand for multimodal systems increases. While generative data augmentation offers a promising solution, existing approaches often produce emotionally inconsistent samples due to oversimplified conditioning on categorical labels. This paper introduces a novel mutual-information-regularised generative framework that combines cross-modal alignment with feature-level synthesis. Building on an InfoGAN-style architecture, our method first learns a semantically aligned audio-text representation space using pre-trained transformers and contrastive objectives. A feature generator is then trained to produce emotion-aware audio features while employing mutual information as a quantitative regulariser to ensure strong dependency between generated features and their conditioning variables. We extend this approach to multimodal settings, enabling the generation of novel, paired (audio, text) features. Comprehensive evaluation on three benchmark datasets (IEMOCAP, MSP-IMPROV, MSP-Podcast) demonstrates that our framework consistently outperforms existing augmentation methods, achieving state-of-the-art performance with improvements of up to 2.6% in unimodal SER and 3.2% in multimodal emotion recognition. Most importantly, we demonstrate that mutual information functions as both a regulariser and a measurable metric for generative quality, offering a systematic approach to data augmentation in affective computing.

The widespread application of large language models (LLMs) raises increasing demands on ensuring safety or imposing constraints, such as reducing harmful content and adhering to predefined rules. While there have been several works studying LLM safety alignment, these works either need to train three models and incur high memory costs, or require prior knowledge on the optimal solution. Witnessing this fact, we investigate the constrained alignment problem for LLMs, i.e., maximizing the reward of outputs while restricting the cost to stay below a threshold. We propose a novel primal-dual direct preference optimization (DPO) approach, which first trains a model using standard DPO on reward preference data to provide reward information, and then adopts a rearranged Lagrangian DPO objective utilizing the provided reward information to fine-tune LLMs. Our approach only needs to train two models rather than three, which significantly saves memory costs, and does not require extra prior knowledge. Moreover, we establish rigorous suboptimality and constraint violation guarantees. We also extend our approach to enable online exploration and drop the data coverage dependence in the results. Experiments on the PKU-SafeRLHF and TruthfulQA datasets demonstrate the state-of-the-art performance of our approach.

Using intelligent systems to perceive psychological and social behaviors, that is, the underlying affective, cognitive, and pathological states that are manifested through observable behaviors and social interactions, remains a challenge due to their complex, multifaceted, and personalized nature. Existing work tackling these dimensions through specialized datasets and single-task systems often miss opportunities for scalability, cross-task transfer, and broader generalization. To address this gap, we curate Human Behavior Atlas, a unified benchmark of diverse behavioral tasks designed to support the development of foundation models for understanding psychological and social behaviors. Human Behavior Atlas comprises over 100,000 samples spanning text, audio, and visual modalities, covering tasks on affective states, cognitive states, pathologies, and social processes. Our unification efforts can reduce redundancy and cost, enable training to scale efficiently across tasks, and enhance generalization of behavioral features across domains. On Human Behavior Atlas, we train three models: Omnisapiens-7B SFT, Omnisapiens-7B BAM, and Omnisapiens-7B RL. We show that training on Human Behavior Atlas enables models to consistently outperform existing multimodal LLMs across diverse behavioral tasks. Pretraining on Human Behavior Atlas also improves transfer to novel behavioral datasets; with the targeted use of behavioral descriptors yielding meaningful performance gains. The benchmark, models, and codes can be found at: https://github.com/MIT-MI/human_behavior_atlas.

Graph foundation models face several fundamental challenges including transferability across datasets and data scarcity, which calls into question the very feasibility of graph foundation models. However, despite similar challenges, the tabular domain has recently witnessed the emergence of the first successful foundation models such as TabPFNv2 and LimiX. Many of these models are based on the prior-data fitted networks (PFN) framework, in which models are pretrained on carefully designed synthetic datasets to make predictions in an in-context learning setting. Recently, G2T-FM has made the first step towards adopting PFNs for graphs, yet it is limited to hand-crafted features and was never pretrained on graph data. In this work, we make the next step by proposing GraphPFN, a PFN-based model designed and pretrained specifically for graph node-level tasks. Following the PFN framework, we first design a prior distribution of synthetic attributed graphs by using a novel combination of multi-level stochastic block models and a preferential attachment process for structure generation and graph-aware structured causal models for attribute generation. Then, we augment the tabular foundation model LimiX with attention-based graph neighborhood aggregation layers and train it on synthetic graphs sampled from our prior. On diverse real-world graph datasets with node-level tasks, GraphPFN shows strong in-context learning performance and achieves state-of-the-art results after finetuning, outperforming both G2T-FM and task-specific GNNs trained from scratch on most datasets. More broadly, GraphPFN shows the potential of PFN-based models for building graph foundation models.

While table understanding increasingly relies on pixel-only settings, current benchmarks predominantly use synthetic renderings that lack the complexity and visual diversity of real-world tables. Additionally, existing visual table understanding (VTU) datasets offer fixed examples with single visualizations and pre-defined instructions, providing no access to underlying serialized data for reformulation. We introduce TABLET, a large-scale VTU dataset with 4 million examples across 21 tasks, grounded in 2 million unique tables where 88% preserve original visualizations. To evaluate whether models are able to jointly reason over tabular and visual content, we also introduce VisualTableQA, a benchmark requiring both visual perception and table understanding. Fine-tuning vision-language models like Qwen2.5-VL-7B and Gemma 3-4B on TABLET improves performance on seen and unseen VTU tasks while increasing robustness on real-world table visualizations. By preserving original visualizations and maintaining example traceability in a unified large-scale collection, TABLET establishes a foundation for robust training and extensible evaluation of future VTU models.

Multi-Head Latent Attention (MLA) is a recent attention mechanism adopted in state-of-the-art LLMs such as DeepSeek-v3 and Kimi K2. Thanks to its novel formulation, MLA allows two functionally equivalent but computationally distinct kernel implementations: naive and absorb. While the naive kernels (e.g., FlashAttention) are typically preferred in training and prefill for their computational efficiency, existing decoding kernels (e.g., FlashMLA) rely on the absorb method to minimize HBM bandwidth usage. However, the compute-bound nature of the absorb implementations prohibits performance benefits from data reuse opportunities in attention calculations, such as shared prefixes. In this work, we introduce TyphoonMLA, a hybrid approach that combines naive and absorb formulations to harness the strengths of both. TyphoonMLA effectively leverages the shared prefix by applying the naive formulation to the compute-bound parts of attention calculations, while reducing the bandwidth requirements for non-shared parts by using the absorb formulation. As a result, TyphoonMLA improves the throughput of attention calculations in MLA architectures by up to 3x and 3.24x on NPU and GPUs, with only a 3% overhead in HBM size.