It is essential for understanding neural network decisions to interpret the functionality (also known as concepts) of neurons. Existing approaches describe neuron concepts by generating natural language descriptions, thereby advancing the understanding of the neural network's decision-making mechanism. However, these approaches assume that each neuron has well-defined functions and provides discriminative features for neural network decision-making. In fact, some neurons may be redundant or may offer misleading concepts. Thus, the descriptions for such neurons may cause misinterpretations of the factors driving the neural network's decisions. To address the issue, we introduce a verification of neuron functions, which checks whether the generated concept highly activates the corresponding neuron. Furthermore, we propose a Select-Hypothesize-Verify framework for interpreting neuron functionality. This framework consists of: 1) selecting activation samples that best capture a neuron's well-defined functional behavior through activation-distribution analysis; 2) forming hypotheses about concepts for the selected neurons; and 3) verifying whether the generated concepts accurately reflect the functionality of the neuron. Extensive experiments show that our method produces more accurate neuron concepts. Our generated concepts activate the corresponding neurons with a probability approximately 1.5 times that of the current state-of-the-art method.

This paper provides some of the first large-scale descriptive evidence on how consumers adopt and use platform-embedded shopping AI in e-commerce. Using data on 31 million users of Ctrip, China's largest online travel platform, we study "Wendao," an LLM-based AI assistant integrated into the platform. We document three empirical regularities. First, adoption is highest among older consumers, female users, and highly engaged existing users, reversing the younger, male-dominated profile commonly documented for general-purpose AI tools. Second, AI chat appears in the same broad phase of the purchase journey as traditional search and well before order placement; among journeys containing both chat and search, the most common pattern is interleaving, with users moving back and forth between the two modalities. Third, consumers disproportionately use the assistant for exploratory, hard-to-keyword tasks: attraction queries account for 42% of observed chat requests, and chat intent varies systematically with both the timing of chat relative to search and the category of products later purchased within the same journey. These findings suggest that embedded shopping AI functions less as a substitute for conventional search than as a complementary interface for exploratory product discovery in e-commerce.

This paper introduces \textbf{FinMCP-Bench}, a novel benchmark for evaluating large language models (LLMs) in solving real-world financial problems through tool invocation of financial model context protocols. FinMCP-Bench contains 613 samples spanning 10 main scenarios and 33 sub-scenarios, featuring both real and synthetic user queries to ensure diversity and authenticity. It incorporates 65 real financial MCPs and three types of samples, single tool, multi-tool, and multi-turn, allowing evaluation of models across different levels of task complexity. Using this benchmark, we systematically assess a range of mainstream LLMs and propose metrics that explicitly measure tool invocation accuracy and reasoning capabilities. FinMCP-Bench provides a standardized, practical, and challenging testbed for advancing research on financial LLM agents.

Vision-Language-Action (VLA) models excel in robotic manipulation but suffer from significant inference latency due to processing dense visual tokens. Existing token reduction methods predominantly rely on attention magnitude as a static selection. In this work, we challenge this assumption, revealing that high-attention tokens are task-dependent and can even degrade policy performance. To address this, we introduce \textbf{TIES} (\textbf{T}au-guided \textbf{I}nter-layer \textbf{E}fficient \textbf{S}election), a dynamic framework guided by inter-layer token ranking consistency. By adaptively balancing attention magnitude with ranking consistency, TIES ensures robust token selection without requiring additional training. On the CogACT + SIMPLER benchmark, TIES improves average success rates by 6\% while reducing token usage by 78\%, and demonstrate strong generalization across diverse decoders and benchmarks.

Predicting human gaze in video is fundamental to advancing scene understanding and multimodal interaction. While traditional saliency maps provide spatial probability distributions and scanpaths offer ordered fixations, both abstractions often collapse the fine-grained temporal dynamics of raw gaze. Furthermore, existing models are typically constrained to short-term windows ($\approx$ 3-5s), failing to capture the long-range behavioral dependencies inherent in real-world content. We present a generative framework for infinite-horizon raw gaze prediction in videos of arbitrary length. By leveraging an autoregressive diffusion model, we synthesize gaze trajectories characterized by continuous spatial coordinates and high-resolution timestamps. Our model is conditioned on a saliency-aware visual latent space. Quantitative and qualitative evaluations demonstrate that our approach significantly outperforms existing approaches in long-range spatio-temporal accuracy and trajectory realism.

We propose Context-aware Video-text Alignment (CVA), a novel framework to address a significant challenge in video temporal grounding: achieving temporally sensitive video-text alignment that remains robust to irrelevant background context. Our framework is built on three key components. First, we propose Query-aware Context Diversification (QCD), a new data augmentation strategy that ensures only semantically unrelated content is mixed in. It builds a video-text similarity-based pool of replacement clips to simulate diverse contexts while preventing the ``false negative" caused by query-agnostic mixing. Second, we introduce the Context-invariant Boundary Discrimination (CBD) loss, a contrastive loss that enforces semantic consistency at challenging temporal boundaries, making their representations robust to contextual shifts and hard negatives. Third, we introduce the Context-enhanced Transformer Encoder (CTE), a hierarchical architecture that combines windowed self-attention and bidirectional cross-attention with learnable queries to capture multi-scale temporal context. Through the synergy of these data-centric and architectural enhancements, CVA achieves state-of-the-art performance on major VTG benchmarks, including QVHighlights and Charades-STA. Notably, our method achieves a significant improvement of approximately 5 points in Recall@1 (R1) scores over state-of-the-art methods, highlighting its effectiveness in mitigating false negatives.

The growing prominence of cryptocurrencies has triggered widespread public engagement and increased speculative activity, particularly on social media platforms. This study introduces a novel classification framework for identifying predictive statements in cryptocurrency-related tweets, focusing on five popular cryptocurrencies: Cardano, Matic, Binance, Ripple, and Fantom. The classification process is divided into two stages: Task 1 involves binary classification to distinguish between Predictive and Non-Predictive statements. Tweets identified as Predictive proceed to Task 2, where they are further categorized as Incremental, Decremental, or Neutral. To build a robust dataset, we combined manual and GPT-based annotation methods and utilized SenticNet to extract emotion features corresponding to each prediction category. To address class imbalance, GPT-generated paraphrasing was employed for data augmentation. We evaluated a wide range of machine learning, deep learning, and transformer-based models across both tasks. The results show that GPT-based balancing significantly enhanced model performance, with transformer models achieving the highest F1-score in Task 1, while traditional machine learning models performed best in Task 2. Furthermore, our emotion analysis revealed distinct emotional patterns associated with each prediction category across the different cryptocurrencies.

Multi-Agent Self-Driving (MASD) systems provide an effective solution for coordinating autonomous vehicles to reduce congestion and enhance both safety and operational efficiency in future intelligent transportation systems. Multi-Agent Reinforcement Learning (MARL) has emerged as a promising approach for developing advanced end-to-end MASD systems. However, achieving efficient and safe collaboration in dynamic MASD systems remains a significant challenge in dense scenarios with complex agent interactions. To address this challenge, we propose a novel collaborative(CO-) interaction-aware(-IN) MARL framework, named COIN. Specifically, we develop a new counterfactual individual-global twin delayed deep deterministic policy gradient (CIG-TD3) algorithm, crafted in a "centralized training, decentralized execution" (CTDE) manner, which aims to jointly optimize the individual objectives (navigation) and the global objectives (collaboration) of agents. We further introduce a dual-level interaction-aware centralized critic architecture that captures both local pairwise interactions and global system-level dependencies, enabling more accurate global value estimation and improved credit assignment for collaborative policy learning. We conduct extensive simulation experiments in dense urban traffic environments, which demonstrate that COIN consistently outperforms other advanced baseline methods in both safety and efficiency across various system sizes. These results highlight its superiority in complex and dynamic MASD scenarios, as further validated through real-world robot demonstrations. Supplementary videos are available at https://marmotlab.github.io/COIN/

Recent advances in robotics, automation, and artificial intelligence have enabled urban traffic systems to operate with increasing autonomy towards future smart cities, powered in part by the development of adaptive traffic signal control (ATSC), which dynamically optimizes signal phases to mitigate congestion and optimize traffic. However, achieving effective and generalizable large-scale ATSC remains a significant challenge due to the diverse intersection topologies and highly dynamic, complex traffic demand patterns across the network. Existing RL-based methods typically use a single shared policy for all scenarios, whose limited representational capacity makes it difficult to capture diverse traffic dynamics and generalize to unseen environments. To address these challenges, we propose CROSS, a novel Mixture-of-Experts (MoE)-based decentralized RL framework for generalizable ATSC. We first introduce a Predictive Contrastive Clustering (PCC) module that forecasts short-term state transitions to identify latent traffic patterns, followed by clustering and contrastive learning to enhance pattern-level representation. We further design a Scenario-Adaptive MoE module that augments a shared policy with multiple experts, thus enabling adaptive specialization and more flexible scenario-specific strategies. We conduct extensive experiments in the SUMO simulator on both synthetic and real-world traffic datasets. Compared with state-of-the-art baselines, CROSS achieves superior performance and generalization through improved representation of diverse traffic scenarios.

Understanding and quantifying uncertainty in large language model (LLM) outputs is critical for reliable deployment. However, traditional evaluation approaches provide limited insight into model confidence at individual token positions during generation. To address this issue, we introduce LogitScope, a lightweight framework for analyzing LLM uncertainty through token-level information metrics computed from probability distributions. By measuring metrics such as entropy and varentropy at each generation step, LogitScope reveals patterns in model confidence, identifies potential hallucinations, and exposes decision points where models exhibit high uncertainty, all without requiring labeled data or semantic interpretation. We demonstrate LogitScope's utility across diverse applications including uncertainty quantification, model behavior analysis, and production monitoring. The framework is model-agnostic, computationally efficient through lazy evaluation, and compatible with any HuggingFace model, enabling both researchers and practitioners to inspect LLM behavior during inference.

Recent work shows that standard greedy-decoding extraction methods for quantifying memorization in LLMs miss how extraction risk varies across sequences. Probabilistic extraction -- computing the probability of generating a target suffix given a prefix under a decoding scheme -- addresses this, but is tractable only for verbatim memorization, missing near-verbatim instances that pose similar privacy and copyright risks. Quantifying near-verbatim extraction risk is expensive: the set of near-verbatim suffixes is combinatorially large, and reliable Monte Carlo (MC) estimation can require ~100,000 samples per sequence. To mitigate this cost, we introduce decoding-constrained beam search, which yields deterministic lower bounds on near-verbatim extraction risk at a cost comparable to ~20 MC samples per sequence. Across experiments, our approach surfaces information invisible to verbatim methods: many more extractable sequences, substantially larger per-sequence extraction mass, and patterns in how near-verbatim extraction risk manifests across model sizes and types of text.

Deploying neural networks on microcontrollers is constrained by kilobytes of flash and SRAM, where 1x1 pointwise (PW) mixers often dominate memory even after INT8 quantization across vision, audio, and wearable sensing. We present HYPER-TINYPW, a compression-as-generation approach that replaces most stored PW weights with generated weights: a shared micro-MLP synthesizes PW kernels once at load time from tiny per-layer codes, caches them, and executes them with standard integer operators. This preserves commodity MCU runtimes and adds only a one-off synthesis cost; steady-state latency and energy match INT8 separable CNN baselines. Enforcing a shared latent basis across layers removes cross-layer redundancy, while keeping PW1 in INT8 stabilizes early, morphology-sensitive mixing. We contribute (i) TinyML-faithful packed-byte accounting covering generator, heads/factorization, codes, kept PW1, and backbone; (ii) a unified evaluation with validation-tuned t* and bootstrap confidence intervals; and (iii) a deployability analysis covering integer-only inference and boot versus lazy synthesis. On three ECG benchmarks (Apnea-ECG, PTB-XL, MIT-BIH), HYPER-TINYPW shifts the macro-F1 versus flash Pareto frontier: at about 225 kB it matches a roughly 1.4 MB CNN while being 6.31x smaller (84.15% fewer bytes), retaining at least 95% of large-model macro-F1. Under 32-64 kB budgets it sustains balanced detection where compact baselines degrade. The mechanism applies broadly to other 1D biosignals, on-device speech, and embedded sensing tasks where per-layer redundancy dominates, indicating a wider role for compression-as-generation in resource-constrained ML systems. Beyond ECG, HYPER-TINYPW transfers to TinyML audio: on Speech Commands it reaches 96.2% test accuracy (98.2% best validation), supporting broader applicability to embedded sensing workloads where repeated linear mixers dominate memory.

Accurately modeling how real-world materials reflect light remains a core challenge in inverse rendering, largely due to the scarcity of real measured reflectance data. Existing approaches rely heavily on synthetic datasets with simplified illumination and limited material realism, preventing models from generalizing to real-world images. We introduce a large-scale polarized reflection and material dataset of real-world objects, captured with an 8-camera, 346-light Light Stage equipped with cross/parallel polarization. Our dataset spans 218 everyday objects across five acquisition dimensions-multiview, multi-illumination, polarization, reflectance separation, and material attributes-yielding over 1.2M high-resolution images with diffuse-specular separation and analytically derived diffuse albedo, specular albedo, and surface normals. Using this dataset, we train and evaluate state-of-the-art inverse and forward rendering models on intrinsic decomposition, relighting, and sparse-view 3D reconstruction, demonstrating significant improvements in material separation, illumination fidelity, and geometric consistency. We hope that our work can establish a new foundation for physically grounded material understanding and enable real-world generalization beyond synthetic training regimes. Project page: https://jingyangcarl.github.io/ICTPolarReal/

Coordinating teams of aerial robots in cluttered three-dimensional (3D) environments requires a principled integration of discrete mission planning-deciding which robot serves which goals and in what order -- with continuous-time trajectory synthesis that enforces collision avoidance and dynamic feasibility. This paper introduces IMD-TAPP (Integrated Multi-Drone Task Allocation and Path Planning), an end-to-end framework that jointly addresses multi-goal allocation, tour sequencing, and safe trajectory generation for quadrotor teams operating in obstacle-rich spaces. IMD--TAPP first discretizes the workspace into a 3D navigation graph and computes obstacle-aware robot-to-goal and goal-to-goal travel costs via graph-search-based pathfinding. These costs are then embedded within an Injected Particle Swarm Optimization (IPSO) scheme, guided by multiple linear assignment, to efficiently explore coupled assignment/ordering alternatives and to minimize mission makespan. Finally, the resulting waypoint tours are transformed into time-parameterized minimum-snap trajectories through a generation-and-optimization routine equipped with iterative validation of obstacle clearance and inter-robot separation, triggering re-planning when safety margins are violated. Extensive MATLAB simulations across cluttered 3D scenarios demonstrate that IMD--TAPP consistently produces dynamically feasible, collision-free trajectories while achieving competitive completion times. In a representative case study with two drones serving multiple goals, the proposed approach attains a minimum mission time of 136~s while maintaining the required safety constraints throughout execution.

We prove that platform-deterministic inference is necessary and sufficient for trustworthy AI. We formalize this as the Determinism Thesis and introduce trust entropy to quantify the cost of non-determinism, proving that verification failure probability equals 1 - 2^{-H_T} exactly. We prove a Determinism-Verification Collapse: verification under determinism requires O(1) hash comparison; without it, the verifier faces an intractable membership problem. IEEE 754 floating-point arithmetic fundamentally violates the determinism requirement. We resolve this by constructing a pure integer inference engine that achieves bitwise identical output across ARM and x86. In 82 cross-architecture tests on models up to 6.7B parameters, we observe zero hash mismatches. Four geographically distributed nodes produce identical outputs, verified by 356 on-chain attestation transactions. Every major trust property of AI systems (fairness, robustness, privacy, safety, alignment) presupposes platform determinism. Our system, 99,000 lines of Rust deployed across three continents, establishes that AI trust is a question of arithmetic.

Accurate surgical phase recognition is essential for analyzing procedural workflows, supporting intraoperative decision-making, and enabling data-driven improvements in surgical education and performance evaluation. In this work, we present a comprehensive framework for phase recognition in pituitary tumor surgery (PTS) videos, combining self-supervised representation learning, robust temporal modeling, and scalable data annotation strategies. Our method achieves 90\% accuracy on a held-out test set, outperforming current state-of-the-art approaches and demonstrating strong generalization across variable surgical cases. A central contribution of this work is the integration of a collaborative online platform designed for surgeons to upload surgical videos, receive automated phase analysis, and contribute to a growing dataset. This platform not only facilitates large-scale data collection but also fosters knowledge sharing and continuous model improvement. To address the challenge of limited labeled data, we pretrain a ResNet-50 model using the self-supervised framework on 251 unlabeled PTS videos, enabling the extraction of high-quality feature representations. Fine-tuning is performed on a labeled dataset of 81 procedures using a modified training regime that incorporates focal loss, gradual layer unfreezing, and dynamic sampling to address class imbalance and procedural variability.

This paper describes LogSigma, our system for SemEval-2026 Task 3: Dimensional Aspect-Based Sentiment Analysis (DimABSA). Unlike traditional Aspect-Based Sentiment Analysis (ABSA), which predicts discrete sentiment labels, DimABSA requires predicting continuous Valence and Arousal (VA) scores on a 1-9 scale. A central challenge is that Valence and Arousal differ in prediction difficulty across languages and domains. We address this using learned homoscedastic uncertainty, where the model learns task-specific log-variance parameters to automatically balance each regression objective during training. Combined with language-specific encoders and multi-seed ensembling, LogSigma achieves 1st place on five datasets across both tracks. The learned variance weights vary substantially across languages due to differing Valence-Arousal difficulty profiles-from 0.66x for German to 2.18x for English-demonstrating that optimal task balancing is language-dependent and cannot be determined a priori.

We investigate machine learning approaches for optimizing real-time staffing decisions in semi-automated warehouse sortation systems. Operational decision-making can be supported at different levels of abstraction, with different trade-offs. We evaluate two approaches, each in a matching simulation environment. First, we train custom Transformer-based policies using offline reinforcement learning on detailed historical state representations, achieving a 2.4% throughput improvement over historical baselines in learned simulators. In high-volume warehouse operations, improvements of this size translate to significant savings. Second, we explore LLMs operating on abstracted, human-readable state descriptions. These are a natural fit for decisions that warehouse managers make using high-level operational summaries. We systematically compare prompting techniques, automatic prompt optimization, and fine-tuning strategies. While prompting alone proves insufficient, supervised fine-tuning combined with Direct Preference Optimization on simulator-generated preferences achieves performance that matches or slightly exceeds historical baselines in a hand-crafted simulator. Our findings demonstrate that both approaches offer viable paths toward AI-assisted operational decision-making. Offline RL excels with task-specific architectures. LLMs support human-readable inputs and can be combined with an iterative feedback loop that can incorporate manager preferences.

Remote sensing visual grounding (RSVG) aims to localize specific targets in remote sensing images using natural language expressions. However, existing methods are restricted to single-sensor domains, i.e., either optical or synthetic aperture radar (SAR), limiting their real-world applicability. In this paper, we introduce the Cross-Domain RSVG (CD-RSVG) task and construct OptSAR-RSVG, the first large-scale benchmark dataset for this setting. To tackle the challenges of cross-domain feature modeling, computational inefficiency, and fine-grained semantic discrimination, we propose OptiSAR-Net++. Our framework features a patch-level Low-Rank Adaptation Mixture of Experts (PL-MoE) for efficient cross-domain feature decoupling. To mitigate the substantial computational overhead of Transformer decoding frameworks, we adopt a CLIP-based contrastive paradigm and further incorporate dynamic adversarial negative sampling, thereby transforming generative regression into an efficient cross-modal matching process. Additionally, a text-guided dual-gate fusion module (TGDF-SSA) and a region-aware auxiliary head are introduced to enhance semantic-visual alignment and spatial modeling. Extensive experiments demonstrate that OptiSAR-Net++ achieves SOTA performance on both OptSAR-RSVG and DIOR-RSVG benchmarks, offering significant advantages in localization accuracy and efficiency. Our code and dataset will be made publicly available.

The physical world is not merely visual; it is governed by rigorous structural and procedural constraints. Yet, the evaluation of vision-language models (VLMs) remains heavily skewed toward perceptual realism, prioritizing the generation of visually plausible 3D layouts, shapes, and appearances. Current benchmarks rarely test whether models grasp the step-by-step processes and physical dependencies required to actually build these artifacts, a capability essential for automating design-to-construction pipelines. To address this, we introduce DreamHouse, a novel benchmark for physical generative reasoning: the capacity to synthesize artifacts that concurrently satisfy geometric, structural, constructability, and code-compliance constraints. We ground this benchmark in residential timber-frame construction, a domain with fully codified engineering standards and objectively verifiable correctness. We curate over 26,000 structures spanning 13 architectural styles, ach verified to construction-document standards (LOD 350) and develop a deterministic 10-test structural validation framework. Unlike static benchmarks that assess only final outputs, DreamHouse supports iterative agentic interaction. Models observe intermediate build states, generate construction actions, and receive structured environmental feedback, enabling a fine-grained evaluation of planning, structural reasoning, and self-correction. Extensive experiments with state-of-the-art VLMs reveal substantial capability gaps that are largely invisible on existing leaderboards. These findings establish physical validity as a critical evaluation axis orthogonal to visual realism, highlighting physical generative reasoning as a distinct and underdeveloped frontier in multimodal intelligence. Available at https://luluyuyuyang.github.io/dreamhouse

Emergency response systems generate data from many agencies and systems. In practice, correlating and updating this information across sources in a way that aligns with Next Generation 9-1-1 data standards remains challenging. Ideally, this data should be treated as a continuous stream of operational updates, where new facts are integrated immediately to provide a timely and unified view of an evolving incident. This paper presents SentinelAI, a data integration and standardization framework for transforming emergency communications into standardized, machine-readable datasets that support integration, composite incident construction, and cross-source reasoning. SentinelAI implements a scalable processing pipeline composed of specialized agents. The EIDO Agent ingests raw communications and produces NENA-compliant Emergency Incident Data Object JSON.

Fire safety consists of a complex pipeline, and it is a very important topic of concern. One of its frontal parts are the smoke detectors, which are supposed to provide an alarm prior to a massive fire appears. As they are often difficult to reach due to high ceilings or problematic locations, an automatic inspection system would be very beneficial as it could allow faster revisions, prevent workers from dangerous work in heights, and make the whole process cheaper. In this study, we present the smoke detector recognition part of the automatic inspection system, which could easily be integrated to the drone system. As part of our research, we compare two popular convolutional-based object detectors YOLOv11 and SSD widely used on embedded devices together with the state-of-the-art transformer-based RT-DETRv2 with the backbones of different sizes. Due to a complicated way of collecting a sufficient amount of data for training in the real-world environment, we also compare several training strategies using the real and semi-synthetic data together with various augmentation methods. To achieve a robust testing, all models were evaluated on two test datasets with an expected and difficult appearance of the smoke detectors including motion blur, small resolution, or not complete objects. The best performing detector is the YOLOv11n, which reaches the average mAP@0.5 score of 0.884. Our code, pretrained models and dataset are publicly available.

Coronary artery disease, the leading cause of cardiovascular mortality worldwide, can be assessed non-invasively by coronary computed tomography angiography (CCTA). Despite progress in automated CCTA analysis using deep learning, clinical translation is constrained by the scarcity of expert-annotated datasets. Furthermore, widely adopted label-free pretraining strategies, such as masked image modeling, are intrinsically biased toward global anatomical statistics, frequently failing to capture the spatially localized pathological features of coronary plaques. Here, we introduce CORA, a 3D vision foundation model for comprehensive cardiovascular risk assessment. CORA learns directly from volumetric CCTA via a pathology-centric, synthesis-driven self-supervised framework. By utilizing an anatomy-guided lesion synthesis engine, the model is explicitly trained to detect simulated vascular abnormalities, biasing representation learning toward clinically relevant disease features rather than dominant background anatomy. We trained CORA on a large-scale cohort of 12,801 unlabeled CCTA volumes and comprehensively evaluated the model across multi-center datasets from nine independent hospitals. Across diagnostic and anatomical tasks, including plaque characterization, stenosis detection, and coronary artery segmentation, CORA consistently outperformed the state-of-the-art 3D vision foundation models, achieving up to a 29\% performance gain. Crucially, by coupling the imaging encoder with a large language model, we extended CORA into a multimodal framework that significantly improved 30-day major adverse cardiac event (MACE) risk stratification. Our results establish CORA as a scalable and extensible foundation for unified anatomical assessment and cardiovascular risk prediction.

Recent advances in multimodal large language models enable new possibilities for image-based decision support. However, their reliability and operational trade-offs in neuroimaging remain insufficiently understood. We present a comprehensive benchmarking study of vision-enabled large language models for 2D neuroimaging using curated MRI and CT datasets covering multiple sclerosis, stroke, brain tumors, other abnormalities, and normal controls. Models are required to generate multiple outputs simultaneously, including diagnosis, diagnosis subtype, imaging modality, specialized sequence, and anatomical plane. Performance is evaluated across four directions: discriminative classification with abstention, calibration, structured-output validity, and computational efficiency. A multi-phase framework ensures fair comparison while controlling for selection bias. Across twenty frontier multimodal models, the results show that technical imaging attributes such as modality and plane are nearly solved, whereas diagnostic reasoning, especially subtype prediction, remains challenging. Tumor classification emerges as the most reliable task, stroke is moderately solvable, while multiple sclerosis and rare abnormalities remain difficult. Few-shot prompting improves performance for several models but increases token usage, latency, and cost. Gemini-2.5-Pro and GPT-5-Chat achieve the strongest overall diagnostic performance, while Gemini-2.5-Flash offers the best efficiency-performance trade-off. Among open-weight architectures, MedGemma-1.5-4B demonstrates the most promising results, as under few-shot prompting, it approaches the zero-shot performance of several proprietary models, while maintaining perfect structured output. These findings provide practical insights into performance, reliability, and efficiency trade-offs, supporting standardized evaluation of multimodal LLMs in neuroimaging.

Given a question, a language model (LM) implicitly encodes a distribution over possible answers. In practice, post-training procedures for LMs often collapse this distribution onto a single dominant mode. While this is generally not a problem for benchmark-style evaluations that assume one correct answer, many real-world tasks inherently involve multiple valid answers or irreducible uncertainty. Examples include medical diagnosis, ambiguous question answering, and settings with incomplete information. In these cases, we would like LMs to generate multiple plausible hypotheses, ideally with confidence estimates for each one, and without computationally intensive repeated sampling to generate non-modal answers. This paper describes a multi-answer reinforcement learning approach for training LMs to perform distributional reasoning over multiple answers during inference. We modify the RL objective to enable models to explicitly generate multiple candidate answers in a single forward pass, internalizing aspects of inference-time search into the model's generative process. Across question-answering, medical diagnostic, and coding benchmarks, we observe improved diversity, coverage, and set-level calibration scores compared to single answer trained baselines. Models trained with our approach require fewer tokens to generate multiple answers than competing approaches. On coding tasks, they are also substantially more accurate. These results position multi-answer RL as a principled and compute-efficient alternative to inference-time scaling procedures such as best-of-k. Code and more information can be found at https://multi-answer-rl.github.io/.

Reinforcement Learning with Verifiable Rewards (RLVR) has significantly advanced the reasoning capabilities of Large Language Models (LLMs). However, methods such as GRPO and DAPO suffer from substantial computational cost, since they rely on sampling many rollouts for each prompt. Moreover, in RLVR the relative advantage is often sparse: many samples become nearly all-correct or all-incorrect, yielding low within-group reward variance and thus weak learning signals. In this paper, we introduce arrol (Accelerating RLVR via online Rollout Pruning), an online rollout pruning method that prunes rollouts during generation while explicitly steering the surviving ones more correctness-balanced to enhance learning signals. Specifically, arrol trains a lightweight quality head on-the-fly to predict the success probability of partial rollouts and uses it to make early pruning decisions. The learned quality head can further weigh candidates to improve inference accuracy during test-time scaling. To improve efficiency, we present a system design that prunes rollouts inside the inference engine and re-batches the remaining ones for log-probability computation and policy updates. Across GRPO and DAPO on Qwen-3 and LLaMA-3.2 models (1B-8B), arrol improves average accuracy by +2.30 to +2.99 while achieving up to 1.7x training speedup, and yielding up to +8.33 additional gains in average accuracy in test-time scaling. The code is available at https://github.com/Hsu1023/ARRoL.

Long-trajectory video generation is a crucial yet challenging task for world modeling primarily due to the limited scalability of existing video diffusion models (VDMs). Autoregressive models, while offering infinite rollout, suffer from visual drift and poor controllability. To address these issues, we propose DCARL, a novel divide-and-conquer, autoregressive framework that effectively combines the structural stability of the divide-and-conquer scheme with the high-fidelity generation of VDMs. Our approach first employs a dedicated Keyframe Generator trained without temporal compression to establish long-range, globally consistent structural anchors. Subsequently, an Interpolation Generator synthesizes the dense frames in an autoregressive manner with overlapping segments, utilizing the keyframes for global context and a single clean preceding frame for local coherence. Trained on a large-scale internet long trajectory video dataset, our method achieves superior performance in both visual quality (lower FID and FVD) and camera adherence (lower ATE and ARE) compared to state-of-the-art autoregressive and divide-and-conquer baselines, demonstrating stable and high-fidelity generation for long trajectory videos up to 32 seconds in length.

We propose a general framework to extend Flow Matching to homogeneous spaces, i.e. quotients of Lie groups. Our approach reformulates the problem as a flow matching task on the underlying Lie group by lifting the data distributions. This strategy avoids the potentially complicated geometry of homogeneous spaces by working directly on Lie groups, which in turn enables us reduce the problem to a Euclidean flow matching task on Lie algebras. In contrast to Riemannian Flow Matching, our method eliminates the need to define and compute premetrics or geodesics, resulting in a simpler, faster, and fully intrinsic framework.

Clinical decisions are high-stakes and require explicit justification, making model interpretability essential for auditing deep clinical models prior to deployment. As the ecosystem of model architectures and explainability methods expands, critical questions remain: Do architectural features like attention improve explainability? Do interpretability approaches generalize across clinical tasks? While prior benchmarking efforts exist, they often lack extensibility and reproducibility, and critically, fail to systematically examine how interpretability varies across the interplay of clinical tasks and model architectures. To address these gaps, we present a comprehensive benchmark evaluating interpretability methods across diverse clinical prediction tasks and model architectures. Our analysis reveals that: (1) attention when leveraged properly is a highly efficient approach for faithfully interpreting model predictions; (2) black-box interpreters like KernelSHAP and LIME are computationally infeasible for time-series clinical prediction tasks; and (3) several interpretability approaches are too unreliable to be trustworthy. From our findings, we discuss several guidelines on improving interpretability within clinical predictive pipelines. To support reproducibility and extensibility, we provide our implementations via PyHealth, a well-documented open-source framework: https://github.com/sunlabuiuc/PyHealth.

Synthetic data generation through document rewriting has emerged as a promising technique for improving language model pretraining, yet most studies focus on English and do not systematically control for the quality of the source data being rewritten. We present a controlled study of how synthetic rewriting interacts with source data quality in the context of Portuguese continued pretraining. Starting from ClassiCC-PT, a Portuguese corpus annotated with STEM and Educational quality scores, we construct two 10B-token subsets at different quality levels and rewrite each into four styles using a 7B instruction-tuned model, producing approximately 40B tokens of synthetic data per condition. We train two English-centric base models (1.1B and 7B parameters) on each condition and evaluate on PoETa V2, a comprehensive 44-task Portuguese benchmark. At the 7B scale, rewriting high-quality data yields a +3.4 NPM gain over the same data unmodified, while rewriting low-quality data provides only +0.5 NPM. At the 1.1B scale, this interaction is weaker, with unmodified low-quality data performing comparably to rewritten high-quality data. Our results demonstrate that synthetic rewriting acts primarily as a quality multiplier rather than a substitute for data curation, and that this effect is scale-dependent.