The extent to which different biological and artificial neural systems rely on equivalent internal representations to support similar tasks remains a central question in neuroscience and machine learning. Prior work typically compares systems using a single representational similarity metric, even though different metrics emphasize distinct facets of representational correspondence. Here we evaluate a suite of representational similarity measures by asking how well each metric recovers known structure across two domains: for artificial models, whether procedurally dissimilar models (differing in architecture or training paradigm) are assigned lower similarity than procedurally matched models; and for neural data, whether responses from distinct cortical regions are separated while responses from the same region align across subjects. Across both vision models and neural recordings, metrics that preserve representational geometry or tuning structure more reliably separate this structure than more flexible mappings such as linear predictivity. To integrate these complementary facets, we adapt Similarity Network Fusion, originally developed for multi-omics integration, to combine similarity graphs across metrics. The resulting fused similarity yields sharper separation of procedurally defined model families and, when applied to neural data, recovers a clearer hierarchical organization of the ventral visual stream that aligns more closely with established anatomical and functional hierarchies than single metrics. Overall, this approach reveals which dimensions of representational correspondence recover meaningful structure in models and brains, and how complementary notions of similarity can be integrated.

Autonomous exploration in complex multi-agent reinforcement learning (MARL) with sparse rewards critically depends on providing agents with effective intrinsic motivation. While artificial curiosity offers a powerful self-supervised signal, it often confuses environmental stochasticity with meaningful novelty. Moreover, existing curiosity mechanisms exhibit a uniform novelty bias, treating all unexpected observations equally. However, peer behavior novelty, which encode latent task dynamics, are often overlooked, resulting in suboptimal exploration in decentralized, communication-free MARL settings. To this end, inspired by how human children adaptively calibrate their own exploratory behaviors via observing peers, we propose a novel approach to enhance multi-agent exploration. We introduce CERMIC, a principled framework that empowers agents to robustly filter noisy surprise signals and guide exploration by dynamically calibrating their intrinsic curiosity with inferred multi-agent context. Additionally, CERMIC generates theoretically-grounded intrinsic rewards, encouraging agents to explore state transitions with high information gain. We evaluate CERMIC on benchmark suites including VMAS, Meltingpot, and SMACv2. Empirical results demonstrate that exploration with CERMIC significantly outperforms SoTA algorithms in sparse-reward environments.

The trajectory data of traffic participants (TPs) is a fundamental resource for evaluating traffic conditions and optimizing policies, especially at urban intersections. Although data acquisition using drones is efficient, existing datasets still have limitations in scene representativeness, information richness, and data fidelity. This study introduces FLUID, comprising a fine-grained trajectory dataset that captures dense conflicts at typical urban signalized intersections, and a lightweight, full-pipeline framework for drone-based trajectory processing. FLUID covers three distinct intersection types, with approximately 5 hours of recording time and featuring over 20,000 TPs across 8 categories. Notably, the dataset records an average of 2.8 vehicle conflicts per minute across all scenes, with roughly 15% of all recorded motor vehicles directly involved in these conflicts. FLUID provides comprehensive data, including trajectories, traffic signals, maps, and raw videos. Comparison with the DataFromSky platform and ground-truth measurements validates its high spatio-temporal accuracy. Through a detailed classification of motor vehicle conflicts and violations, FLUID reveals a diversity of interactive behaviors, demonstrating its value for human preference mining, traffic behavior modeling, and autonomous driving research.

Political language is deeply intertwined with social identities. While social identities are often shaped by specific cultural contexts and expressed through particular uses of language, existing datasets for group and identity detection are predominantly English-centric, single-label and focus on coarse identity categories. We introduce HebID, the first multilabel Hebrew corpus for social identity detection: 5,536 sentences from Israeli politicians' Facebook posts (Dec 2018-Apr 2021), manually annotated for twelve nuanced social identities (e.g. Rightist, Ultra-Orthodox, Socially-oriented) grounded by survey data. We benchmark multilabel and single-label encoders alongside 2B-9B-parameter generative LLMs, finding that Hebrew-tuned LLMs provide the best results (macro-$F_1$ = 0.74). We apply our classifier to politicians' Facebook posts and parliamentary speeches, evaluating differences in popularity, temporal trends, clustering patterns, and gender-related variations in identity expression. We utilize identity choices from a national public survey, enabling a comparison between identities portrayed in elite discourse and the public's identity priorities. HebID provides a comprehensive foundation for studying social identities in Hebrew and can serve as a model for similar research in other non-English political contexts.

While recent flow-based image editing models demonstrate general-purpose capabilities across diverse tasks, they often struggle to specialize in challenging scenarios -- particularly those involving large-scale shape transformations. When performing such structural edits, these methods either fail to achieve the intended shape change or inadvertently alter non-target regions, resulting in degraded background quality. We propose Follow-Your-Shape, a training-free and mask-free framework that supports precise and controllable editing of object shapes while strictly preserving non-target content. Motivated by the divergence between inversion and editing trajectories, we compute a Trajectory Divergence Map (TDM) by comparing token-wise velocity differences between the inversion and denoising paths. The TDM enables precise localization of editable regions and guides a Scheduled KV Injection mechanism that ensures stable and faithful editing. To facilitate a rigorous evaluation, we introduce ReShapeBench, a new benchmark comprising 120 new images and enriched prompt pairs specifically curated for shape-aware editing. Experiments demonstrate that our method achieves superior editability and visual fidelity, particularly in tasks requiring large-scale shape replacement.

Large Language Models (LLMs) have shown impressive performance across various domains, but their ability to perform molecular reasoning remains underexplored. Existing methods mostly rely on general-purpose prompting, which lacks domain-specific molecular semantics, or fine-tuning, which faces challenges in interpretability and reasoning depth, often leading to structural and textual hallucinations. To address these issues, we introduce MolReasoner, a two-stage framework that transitions LLMs from memorization to high-fidelity chemical reasoning. In the Mol-SFT stage, knowledge-enhanced Chain-of-Thought (CoT) data provides a strong foundation, while the Mol-RL stage refines reasoning using a novel, task-adaptive reward system to mitigate hallucinations. Extensive evaluations demonstrate that MolReasoner significantly outperforms a wide range of strong baselines in both molecule generation and captioning tasks. Further analyses highlight the framework's synergistic design and its ability to produce more interpretable outputs. Our work presents a principled and effective new approach for advancing high-fidelity molecular reasoning.

Real-world applications, such as autonomous driving and humanoid robot manipulation, require precise spatial perception. However, it remains underexplored how Vision-Language Models (VLMs) recognize spatial relationships and perceive spatial movement. In this work, we introduce a spatial evaluation pipeline and construct a corresponding benchmark. Specifically, we categorize spatial understanding into two main types: absolute spatial understanding, which involves querying the absolute spatial position (e.g., left, right) of an object within an image, and 3D spatial understanding, which includes movement and rotation. Notably, our dataset is entirely synthetic, enabling the generation of test samples at a low cost while also preventing dataset contamination. We conduct experiments on multiple state-of-the-art VLMs and observe that there is significant room for improvement in their spatial understanding abilities. Explicitly, in our experiments, humans achieve near-perfect performance on all tasks, whereas current VLMs attain human-level performance only on the two simplest tasks. For the remaining tasks, the performance of VLMs is distinctly lower than that of humans. In fact, the best-performing Vision-Language Models even achieve near-zero scores on multiple tasks. The dataset and code are available on https://github.com/kong13661/LRR-Bench.

Graph neural networks (GNNs) have emerged as a powerful framework for a wide range of node-level graph learning tasks. However, their performance typically depends on random or minimally informed initial feature representations, where poor initialization can lead to slower convergence and increased training instability. In this paper, we address this limitation by leveraging a statistically grounded one-hot graph encoder embedding (GEE) as a high-quality, structure-aware initialization for node features. Integrating GEE into standard GNNs yields the GEE-powered GNN (GG) framework. Across extensive simulations and real-world benchmarks, GG provides consistent and substantial performance gains in both unsupervised and supervised settings. For node classification, we further introduce GG-C, which concatenates the outputs of GG and GEE and outperforms competing methods, achieving roughly 10-50% accuracy improvements across most datasets. These results demonstrate the importance of principled, structure-aware initialization for improving the efficiency, stability, and overall performance of graph neural network architecture, enabling models to better exploit graph topology from the outset.

Interpreting Convolutional Neural Networks (CNNs) is critical for safety-sensitive applications such as healthcare and autonomous systems. Popular visual explanation methods like Grad-CAM use a single convolutional layer, potentially missing multi-scale cues and producing unstable saliency maps. We introduce Winsor-CAM, a single-pass gradient-based method that aggregates Grad-CAM maps from all convolutional layers and applies percentile-based Winsorization to attenuate outlier contributions. A user-controllable percentile parameter p enables semantic-level tuning from low-level textures to high-level object patterns. We evaluate Winsor-CAM on six CNN architectures using PASCAL VOC 2012 and PolypGen, comparing localization (IoU, center-of-mass distance) and fidelity (insertion/deletion AUC) against seven baselines including Grad-CAM, Grad-CAM++, LayerCAM, ScoreCAM, AblationCAM, ShapleyCAM, and FullGrad. On DenseNet121 with a subset of Pascal VOC 2012, Winsor-CAM achieves 46.8% IoU and 0.059 CoM distance versus 39.0% and 0.074 for Grad-CAM, with improved insertion AUC (0.656 vs. 0.623) and deletion AUC (0.197 vs. 0.242). Notably, even the worst-performing fixed p-value configuration outperforms FullGrad across all metrics. An ablation study confirms that incorporating earlier layers improves localization. Similar evaluation on PolypGen polyp segmentation further validates Winsor-CAM's effectiveness in medical imaging contexts. Winsor-CAM provides an efficient, robust, and human-tunable explanation tool for expert-in-the-loop analysis.

Molecular dynamics is crucial for understanding molecular systems but its applicability is often limited by the vast timescales of rare events like protein folding. Enhanced sampling techniques overcome this by accelerating the simulation along key reaction pathways, which are defined by collective variables (CVs). However, identifying effective CVs that capture the slow, macroscopic dynamics of a system remains a major bottleneck. This work proposes a novel framework coined BioEmu-CV that learns these essential CVs automatically from BioEmu, a recently proposed foundation model for generating protein equilibrium samples. In particular, we re-purpose BioEmu to learn time-lagged generation conditioned on the learned CV, i.e., predict the distribution of molecular states after a certain amount of time. This training process promotes the CV to encode only the slow, long-term information while disregarding fast, random fluctuations. We validate our learned CV on fast-folding proteins with two key applications: (1) estimating free energy differences using on-the-fly probability enhanced sampling and (2) sampling transition paths with steered molecular dynamics. Our empirical study also serves as a new systematic and comprehensive benchmark for MLCVs on fast-folding proteins larger than Alanine Dipeptide.

Mixture-of-Experts (MoE) enables efficient scaling of large language models by activating only a subset of experts per input token. However, deploying MoE-based models incurs significant memory overhead due to the need to retain all experts in memory. While structured pruning is promising to reduce memory costs, existing methods often show suboptimal performance and unstable degradation in three dimensions: model architectures, calibration data sources, and calibration sample sizes. This paper proposes Mixture-of-Novices-and-Experts (MoNE), a novel expert pruning method that replaces redundant experts with lightweight novices to achieve effective and robust model compression. MoNE evaluates expert redundancy based on two metrics: access frequency and output variance. Experts exhibiting low usage and stable outputs are pruned and replaced with lightweight novices-unbiased estimations of their original outputs-minimizing performance degradation. Extensive experiments demonstrate that MoNE consistently outperforms baseline methods with minimal accuracy degradation across the three dimensions, confirming its effectiveness and robustness. Notably, it outperforms baselines by up to 2.72 for the average zero shot accuracy across nine downstream tasks under 25% pruning ratio, with only 0.14 performance drop for Qwen2-57B-A14B. The code is available at https://github.com/zxgx/mode-pd.

Conducting supervised and preference fine-tuning of large language models (LLMs) requires high-quality datasets to improve their ability to follow instructions and align with human preferences and values. However, constructing such datasets is resource-intensive, and most publicly available datasets are in English. To address these challenges, we propose the \underline{\textbf{Ta}}xonomy-Guided \underline{\textbf{P}}reference Data Generation (TaP) framework for automated, scalable preference dataset construction across languages. TaP uses a structured taxonomy to provide fine-grained control over dataset composition, ensuring diversity and broad coverage. We use TaP-generated datasets to perform supervised and preference fine-tuning on multiple LLMs. Experimental results demonstrate that LLMs trained on TaP-generated datasets outperform those trained on existing open-source datasets. Remarkably, LLMs trained on TaP-generated datasets outperform models trained on an open-source dataset that is 180$\times$ larger.

Medical image diagnosis is challenging because many diseases resemble normal anatomy and exhibit substantial interpatient variability. Clinicians routinely rely on comparative diagnosis, such as referencing cross-patient healthy control images to identify subtle but clinically meaningful abnormalities. Although healthy reference images are abundant in practice, existing medical vision-language models (VLMs) primarily operate in a single-image or single-series setting and lack explicit mechanisms for comparative diagnosis. This work investigates whether incorporating clinically motivated comparison can enhance VLM performance. We show that providing VLMs with both a query image and a matched healthy reference image, accompanied by cross-patient comparative prompts, significantly improves diagnostic performance. This performance can be further augmented by lightweight supervised fine-tuning (SFT) on a small amount of data. At the same time, we evaluate multiple strategies for selecting reference images, including random sampling, demographic attribute matching, embedding-based retrieval, and cross-center selection, and find consistently strong performance across all settings. Finally, we investigate why comparative diagnosis is effective theoretically, and observe improved sample efficiency and tighter alignment between visual and textual representations. Our findings highlight the clinical relevance of comparison-based diagnosis, provide practical strategies for incorporating reference images into VLMs, and demonstrate improved performance across diverse medical imaging tasks.

The safety of autonomous driving systems (ADS) depends on accurate perception across distance and driving conditions. The outputs of AI perception algorithms are stochastic, which have a major impact on decision making and safety outcomes, including time-to-collision estimation. However, current perception evaluation metrics do not reflect the stochastic nature of perception algorithms. We introduce the Perception Characteristics Distance (PCD), a novel metric incorporating model output uncertainty as represented by the farthest distance at which an object can be reliably detected. To represent a system's overall perception capability in terms of reliable detection distance, we average PCD values across multiple detection quality and probabilistic thresholds to produce the average PCD (aPCD). For empirical validation, we present the SensorRainFall dataset, collected on the Virginia Smart Road using a sensor-equipped vehicle (cameras, radar, and LiDAR) under different weather (clear and rainy) and illumination conditions (daylight, streetlight, and nighttime). The dataset includes ground-truth distances, bounding boxes, and segmentation masks for target objects. Experiments with state-of-the-art models show that aPCD captures meaningful differences across weather, daylight, and illumination conditions, which traditional evaluation metrics fail to reflect. PCD provides an uncertainty-aware measure of perception performance, supporting safer and more robust ADS operation, while the SensorRainFall dataset offers a valuable benchmark for evaluation. The SensorRainFall dataset is publicly available at https://www.kaggle.com/datasets/datadrivenwheels/sensorrainfall, and the evaluation code is available at https://github.com/datadrivenwheels/PCD_Python.

Graph retrieval-augmented generation (GraphRAG) has emerged as a powerful paradigm for enhancing large language models (LLMs) with external knowledge. It leverages graphs to model the hierarchical structure between specific concepts, enabling more coherent and effective knowledge retrieval for accurate reasoning.Despite its conceptual promise, recent studies report that GraphRAG frequently underperforms vanilla RAG on many real-world tasks. This raises a critical question: Is GraphRAG really effective, and in which scenarios do graph structures provide measurable benefits for RAG systems? To address this, we propose GraphRAG-Bench, a comprehensive benchmark designed to evaluate GraphRAG models onboth hierarchical knowledge retrieval and deep contextual reasoning. GraphRAG-Bench features a comprehensive dataset with tasks of increasing difficulty, coveringfact retrieval, complex reasoning, contextual summarization, and creative generation, and a systematic evaluation across the entire pipeline, from graph constructionand knowledge retrieval to final generation. Leveraging this novel benchmark, we systematically investigate the conditions when GraphRAG surpasses traditional RAG and the underlying reasons for its success, offering guidelines for its practical application. All related resources and analyses are collected for the community at https://github.com/GraphRAG-Bench/GraphRAG-Benchmark.

Virtual try-on (VTON) has been widely explored for rendering garments onto person images, while its inverse task, virtual try-off (VTOFF), remains largely overlooked. VTOFF aims to recover standardized product images of garments directly from photos of clothed individuals. This capability is of great practical importance for e-commerce platforms, large-scale dataset curation, and the training of foundation models. Unlike VTON, which must handle diverse poses and styles, VTOFF naturally benefits from a consistent output format in the form of flat garment images. However, existing methods face two major limitations: (i) exclusive reliance on visual cues from a single photo often leads to ambiguity, and (ii) generated images usually suffer from loss of fine details, limiting their real-world applicability. To address these challenges, we introduce TEMU-VTOFF, a Text-Enhanced MUlti-category framework for VTOFF. Our architecture is built on a dual DiT-based backbone equipped with a multimodal attention mechanism that jointly exploits image, text, and mask information to resolve visual ambiguities and enable robust feature learning across garment categories. To explicitly mitigate detail degradation, we further design an alignment module that refines garment structures and textures, ensuring high-quality outputs. Extensive experiments on VITON-HD and Dress Code show that TEMU-VTOFF achieves new state-of-the-art performance, substantially improving both visual realism and consistency with target garments.

Top-$k$ decoding is a widely used method for sampling from LLMs: at each token, only the largest $k$ next-token-probabilities are kept, and the next token is sampled after re-normalizing them to sum to unity. Top-$k$ and other sampling methods are motivated by the intuition that true next-token distributions are sparse, and the noisy LLM probabilities need to be truncated. However, to our knowledge, a precise theoretical motivation for the use of top-$k$ decoding is missing. In this work, we develop a theoretical framework that both explains and generalizes top-$k$ decoding. We view decoding at a fixed token as the recovery of a sparse probability distribution. We consider \emph{Bregman decoders} obtained by minimizing a separable Bregman divergence (for both the \emph{primal} and \emph{dual} cases) with a sparsity-inducing $\ell_0$ regularization. Despite the combinatorial nature of the objective, we show how to optimize it efficiently for a large class of divergences. We show that the optimal decoding strategies are greedy, and further that the loss function is discretely convex in $k$, so that binary search provably and efficiently finds the optimal $k$. We show that top-$k$ decoding arises as a special case for the KL divergence, and identify new decoding strategies that have distinct behaviors (e.g., non-linearly up-weighting larger probabilities after re-normalization).

This work investigates the manufacturing of complex shapes parts with wire arc additive manufacturing (WAAM). In order to guarantee the integrity and quality of each deposited layer that composes the final piece, the deposition process is usually carried out in a flat position. However, for complex geometry parts with non-flat surfaces, this strategy causes unsupported overhangs and staircase effect, which contribute to a poor surface finishing. Generally, the build direction is not constant for every deposited section or layer in complex geometry parts. As a result, there is an additional concern to ensure the build direction is aligned with gravity, thus improving the quality of the final part. This paper proposes an algorithm to control the torch motion with respect to a deposition substrate as well as the torch orientation with respect to an inertial frame. The control scheme is based on task augmentation applied to an extended kinematic chain composed by two robots, which constitutes a coordinated control problem, and allows the deposition trajectory to be planned with respect to the deposition substrate coordinate frame while aligning each layer buildup direction with gravity (or any other direction defined for an inertial frame). Parts with complex geometry aspects have been produced in a WAAM cell composed by two robots (a manipulator with a welding torch and a positioning table holding the workpiece) in order to validate the proposed approach.

Ensuring fairness in machine learning models is critical, particularly in high-stakes domains where biased decisions can lead to serious societal consequences. Existing preprocessing approaches generally lack transparent mechanisms for identifying which features or instances are responsible for unfairness. This obscures the rationale behind data modifications. We introduce FairSHAP, a novel pre-processing framework that leverages Shapley value attribution to improve both individual and group fairness. FairSHAP identifies fairness-critical instances in the training data using an interpretable measure of feature importance, and systematically modifies them through instance-level matching across sensitive groups. This process reduces discriminative risk - an individual fairness metric - while preserving data integrity and model accuracy. We demonstrate that FairSHAP significantly improves demographic parity and equality of opportunity across diverse tabular datasets, achieving fairness gains with minimal data perturbation and, in some cases, improved predictive performance. As a model-agnostic and transparent method, FairSHAP integrates seamlessly into existing machine learning pipelines and provides actionable insights into the sources of bias.Our code is on https://github.com/ZhuMuMu0216/FairSHAP.

Accurately identifying the synthesis conditions of metal-organic frameworks (MOFs) is essential for guiding experimental design, yet remains challenging because relevant information in the literature is often scattered, inconsistent, and difficult to interpret. We present MOFh6, a large language model driven system that reads raw articles or crystal codes and converts them into standardized synthesis tables. It links related descriptions across paragraphs, unifies ligand abbreviations with full names, and outputs structured parameters ready for use. MOFh6 achieved 99% extraction accuracy, resolved 94.1% of abbreviation cases across five major publishers, and maintained a precision of 0.93 +/- 0.01. Processing a full text takes 9.6 s, locating synthesis descriptions 36 s, with 100 papers processed for USD 4.24. By replacing static database lookups with real-time extraction, MOFh6 reshapes MOF synthesis research, accelerating the conversion of literature knowledge into practical synthesis protocols and enabling scalable, data-driven materials discovery.

We model human and animal learning by computing with high-dimensional vectors (H = 10,000 for example). The architecture resembles traditional (von Neumann) computing with numbers, but the instructions refer to vectors and operate on them in superposition. The architecture includes a high-capacity memory for vectors, analogue of the random-access memory (RAM) for numbers. The model's ability to learn from data reminds us of deep learning, but with an architecture closer to biology. The architecture agrees with an idea from psychology that human memory and learning involve a short-term working memory and a long-term data store. Neuroscience provides us with a model of the long-term memory, namely, the cortex of the cerebellum. With roots in psychology, biology, and traditional computing, a theory of computing with vectors can help us understand how brains compute. Application to learning by robots seems inevitable, but there is likely to be more, including language. Ultimately we want to compute with no more material and energy than used by brains. To that end, we need a mathematical theory that agrees with psychology and biology, and is suitable for nanotechnology. We also need to exercise the theory in large-scale experiments. Computing with vectors is described here in terms familiar to us from traditional computing with numbers.

AI models have made significant strides in recent years in their ability to describe and answer questions about real-world images. They have also made progress in the ability to converse with users in real-time using audio input. This raises the question: have we reached the point where AI models, connected to a camera and microphone, can converse with users in real-time about scenes and events that are unfolding live in front of the camera? This has been a long-standing goal in AI and is a prerequisite for real-world AI assistants and humanoid robots to interact with humans in everyday situations. In this work, we introduce a new dataset and benchmark, the Qualcomm Interactive Video Dataset (IVD), which allows us to assess the extent to which existing models can support these abilities, and to what degree these capabilities can be instilled through fine-tuning. The dataset is based on a simple question-answering setup, where users ask questions that the system has to answer, in real-time, based on the camera and audio input. We show that existing models fall far behind human performance on this task, and we identify the main sources for the performance gap. However, we also show that for many of the required perceptual skills, fine-tuning on this form of data can significantly reduce this gap.

We propose V-Droid, a mobile GUI task automation agent. Unlike previous mobile agents that utilize Large Language Models (LLMs) as generators to directly generate actions at each step, V-Droid employs LLMs as verifiers to evaluate candidate actions before making final decisions. To realize this novel paradigm, we introduce a comprehensive framework for constructing verifier-driven mobile agents: the discretized action space construction coupled with the prefilling-only workflow to accelerate the verification process, the pair-wise progress preference training to significantly enhance the verifier's decision-making capabilities, and the scalable human-agent joint annotation scheme to efficiently collect the necessary data at scale. V-Droid obtains a substantial task success rate across several public mobile task automation benchmarks: 59.5% on AndroidWorld, 38.3% on AndroidLab, and 49% on MobileAgentBench, surpassing existing agents by 5.2%, 2.1%, and 9%, respectively. Furthermore, V-Droid achieves a remarkably low latency of 4.3s per step, which is 6.1x faster compared with existing mobile agents. The source code is available at https://github.com/V-Droid-Agent/V-Droid.

Videos, with their unique temporal dimension, demand precise grounded understanding, where answers are directly linked to visual, interpretable evidence. Despite significant breakthroughs in text-based reasoning with large language models, multi-modal reasoning - especially for videos - remains limited. In this work, we fill this gap by introducing VideoMind, a novel video-language agent for temporal-grounded video reasoning. Our method involves two key innovations: (1) We identify four essential capabilities for grounded video reasoning and propose a role-based agentic workflow, comprising a planner to coordinate roles, a grounder for temporal event localization, a verifier to assess event candidates, and an answerer for question answering. (2) To efficiently integrate these roles during inference, we propose a novel Chain-of-LoRA mechanism, where a unified base model with multiple LoRA adapters is leveraged to enable seamless role switching, balancing efficiency and flexibility. Extensive experiments on 15 benchmarks across Grounded VideoQA, Video Temporal Grounding, and General VideoQA tasks demonstrate the effectiveness of the proposed scheme in advancing video agent, test-time scaling, and long-form video reasoning. Code, models, datasets, and demos are available at https://videomind.github.io/.

Test-time training (TTT) methods explicitly update the weights of a model to adapt to the specific test instance, and they have found success in a variety of settings, including most recently language modeling and reasoning. To demystify this success, we investigate a gradient-based TTT algorithm for in-context learning, where we train a transformer model on the in-context demonstrations provided in the test prompt. Specifically, we provide a comprehensive theoretical characterization of linear transformers when the update rule is a single gradient step. Our theory (i) delineates the role of alignment between pretraining distribution and target task, (ii) demystifies how TTT can alleviate distribution shift, and (iii) quantifies the sample complexity of TTT including how it can significantly reduce the eventual sample size required for in-context learning. As our empirical contribution, we study the benefits of TTT for TabPFN, a tabular foundation model. In line with our theory, we demonstrate that TTT significantly reduces the required sample size for tabular classification (3 to 5 times fewer) unlocking substantial inference efficiency with a negligible training cost.

The reliance on Deep Neural Network (DNN)-based classifiers in safety-critical and real-world applications necessitates Open-Set Recognition (OSR). OSR enables the identification of input data from classes unknown during training as unknown, as opposed to misclassifying them as belonging to a known class. DNNs consist of a feature extraction backbone and classifier head; however, most OSR methods typically train both components jointly, often yielding feature representations that adapt poorly to unknown data. Other approaches employ off-the-shelf objectives, such as supervised contrastive learning, which are not specifically designed for OSR. To address these limitations, we propose SpHOR, which explicitly shapes the feature space via supervised representation learning, before training a classifier. Instead of relying on generic feature learning, SpHOR custom-designs representation learning for OSR through three key innovations: (1) enforcing discriminative class-specific features via orthogonal label embeddings, ensuring clearer separation between classes. (2) imposing a spherical constraint, modeling representations as a mixture of von Mises-Fisher distributions. (3) integrating Mixup and Label Smoothing (LS) directly into the representation learning stage. To quantify how these techniques enhance representations for OSR, we introduce two metrics: the Angular Separability (AS) and Norm Separability (NS). Combining all three innovations, SpHOR achieves state-of-the-art results (in AUROC and OSCR) across various coarse-grained and fine-grained open-set benchmarks, particularly excelling on the Semantic Shift Benchmark with improvements up to 5.1\%. Code at https://github.com/nadarasarbahavan/SpHOR

Clearsky models are widely used in solar energy for many applications such as quality control, resource assessment, satellite-base irradiance estimation and forecasting. However, their use in forecasting and nowcasting is associated with a number of challenges. Synchronization errors, reliance on the Clearsky index (ratio of the global horizontal irradiance to its cloud-free counterpart) and high sensitivity of the clearsky model to errors in aerosol optical depth at low solar elevation limit their added value in real-time applications. This paper explores the feasibility of short-term forecasting without relying on a clearsky model. We propose a Clearsky-Free forecasting approach using Extreme Learning Machine (ELM) models. ELM learns daily periodicity and local variability directly from raw Global Horizontal Irradiance (GHI) data. It eliminates the need for Clearsky normalization, simplifying the forecasting process and improving scalability. Our approach is a non-linear adaptative statistical method that implicitely learns the irradiance in cloud-free conditions removing the need for an clear-sky model and the related operational issues. Deterministic and probabilistic results are compared to traditional benchmarks, including ARMA with McClear-generated Clearsky data and quantile regression for probabilistic forecasts. ELM matches or outperforms these methods, providing accurate predictions and robust uncertainty quantification. This approach offers a simple, efficient solution for real-time solar forecasting. By overcoming the stationarization process limitations based on usual multiplicative scheme Clearsky models, it provides a flexible and reliable framework for modern energy systems.

Visual prompt tuning offers significant advantages for adapting pre-trained visual foundation models to specific tasks. However, current research provides limited insight into the interpretability of this approach, which is essential for enhancing AI reliability and enabling AI-driven knowledge discovery. In this paper, rather than learning abstract prompt embeddings, we propose the first framework, named Interpretable Visual Prompt Tuning (IVPT), to explore interpretability for visual prompts by introducing cross-layer concept prototypes. Specifically, visual prompts are linked to human-understandable semantic concepts, represented as a set of category-agnostic prototypes, each corresponding to a specific region of the image. IVPT then aggregates features from these regions to generate interpretable prompts for multiple network layers, allowing the explanation of visual prompts at different network depths and semantic granularities. Comprehensive qualitative and quantitative evaluations on fine-grained classification benchmarks show its superior interpretability and performance over visual prompt tuning methods and existing interpretable methods.

AI agents are increasingly being deployed to automate tasks, often based on underspecified user instructions. Making unwarranted assumptions to compensate for the missing information and failing to ask clarifying questions can lead to suboptimal outcomes, safety risks due to tool misuse, and wasted computational resources. In this work, we study the ability of LLM agents to handle underspecified instructions in interactive code generation settings by evaluating proprietary and open-weight models on their performance across three key steps: (a) detecting underspecificity, (b) asking targeted clarification questions, and (c) leveraging the interaction to improve performance in underspecified scenarios. We introduce Ambig-SWE, an underspecified variant of SWE-Bench Verified, specifically designed to evaluate agent behavior under ambiguity and interaction. Our findings reveal that models struggle to distinguish between well-specified and underspecified instructions. However, when models interact for underspecified inputs, they effectively obtain vital information from the user leading to significant improvements in performance, up to 74% over the non-interactive settings, underscoring the value of effective interaction. Our study highlights critical gaps in how current state-of-the-art models handle missing information in complex software engineering tasks and structures the evaluation into distinct steps to enable targeted improvements.

Oversmoothing is a fundamental challenge in graph neural networks (GNNs): as the number of layers increases, node embeddings become increasingly similar, and model performance drops sharply. Traditionally, oversmoothing has been quantified using metrics that measure the similarity of neighbouring node features, such as the Dirichlet energy. We argue that these metrics have critical limitations and fail to reliably capture oversmoothing in realistic scenarios. For instance, they provide meaningful insights only for very deep networks, while typical GNNs show a performance drop already with as few as 10 layers. As an alternative, we propose measuring oversmoothing by examining the numerical or effective rank of the feature representations. We provide extensive numerical evaluation across diverse graph architectures and datasets to show that rank-based metrics consistently capture oversmoothing, whereas energy-based metrics often fail. Notably, we reveal that drops in the rank align closely with performance degradation, even in scenarios where energy metrics remain unchanged. Along with the experimental evaluation, we provide theoretical support for this approach, clarifying why Dirichlet-like measures may fail to capture performance drop and proving that the numerical rank of feature representations collapses to one for a broad family of GNN architectures.