The demand for efficient large language model (LLM) inference has intensified the focus on sparsification techniques. While semi-structured (N:M) pruning is well-established for weights, its application to activation pruning remains underexplored despite its potential for dynamic, input-adaptive compression and reductions in I/O overhead. This work presents a comprehensive analysis of methods for post-training N:M activation pruning in LLMs. Across multiple LLMs, we demonstrate that pruning activations enables superior preservation of generative capabilities compared to weight pruning at equivalent sparsity levels. We evaluate lightweight, plug-and-play error mitigation techniques and pruning criteria, establishing strong hardware-friendly baselines that require minimal calibration. Furthermore, we explore sparsity patterns beyond NVIDIA's standard 2:4, showing that the 16:32 pattern achieves performance nearly on par with unstructured sparsity. However, considering the trade-off between flexibility and hardware implementation complexity, we focus on the 8:16 pattern as a superior candidate. Our findings provide both effective practical methods for activation pruning and a motivation for future hardware to support more flexible sparsity patterns. Our code is available https://anonymous.4open.science/r/Structured-Sparse-Activations-Inference-EC3C/README.md .

In modern GPU inference, cache efficiency remains a major bottleneck, and heuristic policies such as \textsc{LRU} can perform far worse than the offline optimum. Existing learning-based caching systems improve hit rates mainly through predictor design, but often follow learned predictions blindly, making performance unreliable when predictions are inaccurate. In contrast, emerging learning-augmented caching algorithms~\cite{pmlr-v80-lykouris18a,mitzenmacher2022algorithms} provide performance guarantees by carefully integrating predictions into caching policies, achieving both \emph{consistency} (near-optimality under perfect predictions) and \emph{robustness} (bounded worst-case performance under prediction errors). However, deployment remains challenging. A practical algorithm should satisfy strict time and space efficiency constraints, which some theoretical work overlooks, while also incurring low deployment overhead. We propose learning-augmented LRU, a deployment-oriented learning-augmented caching algorithm that guarantees \emph{1-consistency} and \emph{$O(k)$-robustness}, incurs low time and space overhead, and maintains strong compatibility. We further build a GPU cache, called \textsc{LCR}, on top of learning-augmented LRU to benefit from its theoretical guarantees and translate them into practical performance. In experiments, \textsc{LCR} reduces P99 time-to-first-token (TTFT) by up to 28.3\% on LLM workloads and increases throughput by up to 24.2\% on deep learning recommendation (DLRM) workloads. Even with poor predictions, performance degrades gracefully and remains close to \textsc{LRU}, demonstrating robustness with practical value.

Video sequences often contain structured noise and background artifacts that obscure dynamic content, posing challenges for accurate analysis and restoration. Robust principal component methods address this by decomposing data into low-rank and sparse components. Still, the sparsity assumption often fails to capture the rich variability present in real video data. To overcome this limitation, a hybrid framework that integrates low-rank temporal modeling with diffusion posterior sampling is proposed. The proposed method, Nuclear Diffusion, is evaluated on a real-world medical imaging problem, namely cardiac ultrasound dehazing, and demonstrates improved dehazing performance compared to traditional RPCA concerning contrast enhancement (gCNR) and signal preservation (KS statistic). These results highlight the potential of combining model-based temporal models with deep generative priors for high-fidelity video restoration.

We consider the problem of delivering multiple packages from a single depot to distinct goal locations using a homogeneous fleet of robots with limited carrying capacity. We propose VCST-RCP, a Voronoi-Constrained Steiner Tree Relay Coordination Planning framework that explicitly treats inter-robot relays as a design primitive. The approach operates in two stages: (i) constructing a sparse relay backbone by combining Voronoi-derived exchange interfaces with Steiner tree optimization, and (ii) synthesizing robot-level pickup, relay, and delivery schedules under capacity and service-time constraints. Unlike traditional methods that rely on direct source-to-destination transport, our framework organizes package flow through a shared relay network, reducing redundant long-haul motion. Extensive experiments across multiple scales show that VCST-RCP reduces total fleet travel distance by an average of 31% (up to nearly 50%) compared to Hungarian assignment and significantly outperforms OR-Tools CVRP, with statistically significant improvements (p < 10^{-3}). These gains translate into over 50% higher delivery efficiency (packages per kilometer), directly improving energy utilization. An ablation study further reveals that optimizing relay placement yields substantially larger improvements than adapting spatial partitioning alone, establishing relay design as the dominant factor governing system performance. Overall, the results demonstrate that relay-based coordination provides a scalable and effective framework for energy-aware multi-robot delivery in real-world logistics settings.

We consider federated learning of linearly-parameterized nonlinear systems. We establish theoretical guarantees on the effectiveness of federated nonlinear system identification compared to centralized approaches, demonstrating that the convergence rate improves as the number of clients increases. Although the convergence rates in the linear and nonlinear cases differ only by a constant, this constant depends on the feature map $ϕ$, which can be carefully chosen in the nonlinear setting to increase excitation and improve performance. We experimentally validate our theory in physical settings where client devices are driven by i.i.d. control inputs and control policies exhibiting i.i.d. random perturbations, ensuring non-active exploration. Experiments use trajectories from nonlinear dynamical systems characterized by real-analytic feature functions, including polynomial and trigonometric components, representative of physical systems including pendulum and quadrotor dynamics. We analyze the convergence behavior of the proposed method under varying noise levels and data distributions. Results show that federated learning consistently improves convergence of any individual client as the number of participating clients increases.

Personalization is crucial for enhancing both the effectiveness and user satisfaction of language technologies, particularly in information-seeking tasks like question answering. Current approaches for personalizing large language models (LLMs) often rely on retrieval-augmented generation (RAG), followed by reinforcement learning with scalar reward signals to teach models how to use retrieved personal context. We believe that these scalar rewards sometimes provide weak, non-instructive feedback, limiting learning efficiency and personalization quality. We introduce VAC, a novel framework for personalized response generation that replaces scalar rewards with natural language feedback (NLF) that are generated conditioned on the user profiles and the question narratives. NLF serves as a rich and actionable supervision signal, allowing the policy model to iteratively refine its outputs and internalize effective personalization strategies. Training alternates between optimizing the feedback model and fine-tuning the policy model on the improved responses, resulting in a policy model that no longer requires feedback at inference. Evaluation on the LaMP-QA benchmark that consists of three diverse domains demonstrates consistent and significant improvements over the state-of-the-art results. Human evaluations further confirm the superior quality of the generated responses. These results demonstrate that NLF provides more effective signals for optimizing personalized question answering.

Type 1 diabetes (T1D) is an autoimmune disorder that leads to the destruction of insulin-producing cells, resulting in insulin deficiency, as to why the affected individuals depend on external insulin injections. However, insulin can decrease blood glucose levels and can cause hypoglycemia. Hypoglycemia is a severe event of low blood glucose levels ($\le$70 mg/dL) with dangerous side effects of dizziness, coma, or death. Data analysis can significantly enhance diabetes care by identifying personal patterns and trends leading to adverse events. Especially, machine learning (ML) models can predict glucose levels and provide early alarms. However, diabetes and hypoglycemia research is limited by the unavailability of large datasets. Thus, this work systematically integrates 15 datasets to provide a large database of 2510 subjects with glucose measurements recorded every 5 minutes. In total, 149 million measurements are included, of which 4% represent values in the hypoglycemic range. Moreover, two sub-databases are extracted. Sub-database I includes demographics, and sub-database II includes heart rate data. The integrated dataset provides an equal distribution of sex and different age levels. As a further contribution, data quality is assessed, revealing that data imbalance and missing values present a significant challenge. Moreover, a correlation study on glucose levels and heart rate data is conducted, showing a relation between 15 and 55 minutes before hypoglycemia.

Uncovering hidden symbolic laws from time series data, as an aspiration dating back to Kepler's discovery of planetary motion, remains a core challenge in scientific discovery and artificial intelligence. While Large Language Models show promise in structured reasoning tasks, their ability to infer interpretable, context-aligned symbolic structures from time series data is still underexplored. To systematically evaluate this capability, we introduce SymbolBench, a comprehensive benchmark designed to assess symbolic reasoning over real-world time series across three tasks: multivariate symbolic regression, Boolean network inference, and causal discovery. Unlike prior efforts limited to simple algebraic equations, SymbolBench spans a diverse set of symbolic forms with varying complexity. We further propose a unified framework that integrates LLMs with genetic programming to form a closed-loop symbolic reasoning system, where LLMs act both as predictors and evaluators. Our empirical results reveal key strengths and limitations of current models, highlighting the importance of combining domain knowledge, context alignment, and reasoning structure to improve LLMs in automated scientific discovery. https://github.com/nuuuh/SymbolBench.

This paper introduces two quasi-metrics for performance assessment of multi-object tracking (MOT) algorithms. One quasi-metric is an extension of the generalised optimal subpattern assignment (GOSPA) metric and measures the discrepancy between sets of objects. The other quasi-metric is an extension of the trajectory GOSPA (T-GOSPA) metric and measures the discrepancy between sets of trajectories. Similar to the GOSPA-based metrics, these quasi-metrics include costs for localisation error for properly detected objects, the number of false objects and the number of missed objects. The T-GOSPA quasi-metric also includes a track switching cost. Differently from the GOSPA and T-GOSPA metrics, the proposed quasi-metrics have the flexibility of penalising missed and false objects with different costs, and the localisation costs are not required to be symmetric. We also explain how to obtain similarity score functions based on these quasi-metrics. The performance of several Bayesian MOT algorithms is assessed with the T-GOSPA quasi-metric via simulations.

Electrocardiograms (ECGs) provide non-invasive measurements of heart activity and are established tools for detecting cardiac arrhythmias. Although supervised machine learning has emerged as a promising approach for automated heartbeat classification, substantial variations in ECG signals across individuals and leads, combined with inconsistent labeling standards and dataset biases, make it difficult to develop generalizable models. Dimensionality reduction maps high-dimensional data into a lower-dimensional space while preserving the underlying structure, enabling visualization and pattern discovery. Conventional methods, e.g., principal component analysis, prioritize large variances and typically overlook subtle yet clinically relevant patterns. Here, we show that nonlinear dimensionality reduction (NLDR) algorithms, e.g., t-SNE and UMAP, can identify medically relevant features in ECG signals without pretraining or prior information. Using the MIT-BIH Arrhythmia Database, we show that: a) applying NLDR to a mixed population of heartbeats reveals inter-individual morphological differences, as signals from the same person cluster together in latent spaces; and b) applying NLDR to heartbeats of a single individual separates normal beats from arrhythmias into distinct clusters, identifiable in an unsupervised manner. To our knowledge, this is the first systematic evaluation of NLDR for unsupervised arrhythmia detection. Both UMAP and t-SNE achieved trustworthiness scores >=0.95, indicating that local neighborhoods are well preserved in the embedding. Classification on 2D embeddings outperforms the original high-dimensional space, with a k-NN classifier discriminating individual recordings with >=80% accuracy and identifying arrhythmias with median accuracy >=98% and median F1-score >=85%. These results show that NLDR holds much promise for cardiac monitoring and personalized healthcare.

Hidden Markov Models (HMMs) are foundational tools for modeling sequential data with latent Markovian structure, yet fitting them to real-world data remains computationally challenging. In this work, we show that pre-trained large language models (LLMs) can effectively model data generated by HMMs via in-context learning (ICL)$\unicode{x2013}$their ability to infer patterns from examples within a prompt. On a diverse set of synthetic HMMs, LLMs achieve predictive accuracy approaching the theoretical optimum. We uncover novel scaling trends influenced by HMM properties, and offer theoretical conjectures for these empirical observations. We also provide practical guidelines for scientists on using ICL as a diagnostic tool for complex data. On real-world animal decision-making tasks, ICL achieves competitive performance with models designed by human experts. To our knowledge, this is the first demonstration that ICL can learn and predict HMM-generated sequences$\unicode{x2013}$an advance that deepens our understanding of in-context learning in LLMs and establishes its potential as a powerful tool for uncovering hidden structure in complex scientific data.

Large Language Models (LLMs) have demonstrated remarkable capabilities in various reasoning-intensive tasks. However, these models exhibit unexpected brittleness, often failing on simple variations of the same underlying task. Existing robustness evaluations predominantly rely on hand-crafted templates or a limited set of perturbation rules. Consequently, such approaches lack the adaptability to probe latent vulnerabilities unique to specific models and remain susceptible to data contamination. To address this, we propose the Math Stress Tester (MaSTer), an automated framework inspired by software stress testing. MaSTer generates adversarial variants via a multi-round rewrite-verify loop, ensuring semantic consistency while successfully inducing model failure. Our framework generates benchmark variants dynamically for each LLM, thus minimizing the risk of data contamination. Experiments on GSM8K and MATH-500 demonstrate the effectiveness of MaSTer on mathematical tasks. Additionally, we validate the framework's extensibility to non-mathematical tasks, highlighting its broad applicability. Furthermore, we demonstrate that the synthesized variants generated by MaSTer can be utilized as a fine-tuning dataset to significantly enhance the model's robustness.

Efficient reproduction of research papers is pivotal to accelerating scientific progress. However, the increasing complexity of proposed methods often renders reproduction a labor-intensive endeavor, necessitating profound domain expertise. To address this, we introduce the paper lineage, which systematically mines implicit knowledge from the cited literature. This algorithm serves as the backbone of our proposed \ours, a multi-agent framework designed to autonomously reproduce experimental code in a complete, end-to-end manner. To ensure code executability, \ours incorporates a sampling-based unit testing strategy for rapid validation. To assess reproduction capabilities, we introduce \ourbench, a benchmark featuring verified implementations, alongside comprehensive metrics for evaluating both reproduction and execution fidelity. Extensive evaluations on PaperBench and \ourbench demonstrate that \ours consistently surpasses existing baselines across all metrics. Notably, it yields substantial improvements in reproduction fidelity and final execution performance. The code is available at https://github.com/AI9Stars/AutoReproduce.

Mixture-of-Experts (MoE) models offer dynamic computation, but are typically deployed as static full-capacity models, missing opportunities for deployment-specific specialization. We introduce PreMoE, a training-free framework that proactively compiles sparse MoE variants for targeted deployment scenarios. At its core is Predicted Expert Utility (PEU), a robust metric for estimating expert importance from router logits through high-confidence threshold filtering and logit transformation, which together stabilize utility estimation under aggressive sparsity. Using PEU scores computed on a small calibration set, PreMoE produces domain-aware expert rankings that can be used to compile either domain-specific specialists or high-efficiency multi-domain generalists, without any retraining. Across MoE models ranging from 30B to 718B parameters, PreMoE achieves up to 50\% sparsity with nearly no performance loss. It further exposes a practical deployment trade-off: specialists maximize in-domain efficiency, while synthesized generalists retain broader cross-domain capability at the same sparsity budget.

Often, multilingual language models are trained with the objective to map semantically similar content (in different languages) in the same latent space. In this paper, we show a nuance in this training objective, and find that by changing the language of the input query, we can improve the question answering ability of language models. We make two main contributions. First, we introduce the term Language Specific Knowledge (LSK) to denote queries that are best answered in an ``expert language'' for a given LLM, thereby enhancing its question-answering ability. We introduce the problem of language selection -- for some queries, language models can perform better when queried in languages other than English, sometimes even better in low-resource languages -- and the goal is to select the optimal language for the query. Second, we introduce a variety of simple to strong baselines to empirically motivate the language selection problem (including one of our own methods called LSKExtractor). During our evaluation, we employ three datasets that contain knowledge about both cultural and social behavioral norms. Overall, the results show that principled language selection can improve the performance of a language model, and that the expected question-to-language map is not always intuitive: Gemma models know most about China and Middle East in Spanish; Qwen models know most about authority and responsibility in Arabic and Chinese. Broadly, our research contributes to the open-source development of language models that are inclusive and more aligned with the cultural and linguistic contexts in which they are deployed.

Tibetan is a low-resource language with minimal parallel speech corpora spanning its three major dialects-Ü-Tsang, Amdo, and Kham-limiting progress in speech modeling. To address this issue, we propose FMSD-TTS, a few-shot, multi-speaker, multi-dialect text-to-speech framework that synthesizes parallel dialectal speech from limited reference audio and explicit dialect labels. Our method features a novel speaker-dialect fusion module and a Dialect-Specialized Dynamic Routing Network (DSDR-Net) to capture fine-grained acoustic and linguistic variations across dialects while preserving speaker identity. Extensive objective and subjective evaluations demonstrate that FMSD-TTS significantly outperforms baselines in both dialectal expressiveness and speaker similarity. We further validate the quality and utility of the synthesized speech through a challenging speech-to-speech dialect conversion task. Our contributions include: (1) a novel few-shot TTS system tailored for Tibetan multi-dialect speech synthesis, (2) the public release of a large-scale synthetic Tibetan speech corpus generated by FMSD-TTS, and (3) an open-source evaluation toolkit for standardized assessment of speaker similarity, dialect consistency, and audio quality.

Entropic measures like Shannon entropy (SE), its quantum mechanical analogue von Neumann entropy, and Kullback-Leibler divergence (KL) are key components in many tools used in physics, information theory, machine learning (ML) and quantum computing. Besides of the significant amounts of SE and KL computations required in these fields, the singularity of their gradients near zero is one of the central mathematical reason inducing the high cost, frequently low robustness and slow convergence of computational tools that rely on these concepts. Here we propose the Fast Entropic Approximations (FEA) - non-singular rational approximations of SE and symmetrized KL, that preserve their main mathematical properties and achieve a mean absolute errors of around $10^-3$ ($10-20$ times better than comparable state-of-the-art computational approximations). We show that FEA allows up to around 2 times faster computation of SE and up to 37 times faster computation of symmetrized KL: it requires only $5$ to $7$ elementary computational operations, as compared to the tens of elementary operations behind SE and KL evaluations based on approximate logarithm schemes with table look-ups, bitshifts, or series approximations. On a set of common benchmarks for the feature selection problem in machine learning, we show that the combined effect of fewer elementary operations, low approximation error, preservation of main mathematical properties, and non-singular gradients allows much faster training of significantly-better models. We demonstrate that FEA enables ML feature extraction that is three orders of magnitude faster, and better in quality then the very popular LASSO feature extraction.

Despite substantial advancements in aligning large language models (LLMs) with human values, current safety mechanisms remain susceptible to jailbreak attacks. We hypothesize that this vulnerability stems from distributional discrepancies between alignment-oriented prompts and malicious prompts. To investigate this, we introduce LogiBreak, a novel and universal black-box jailbreak method that leverages logical expression translation to circumvent LLM safety systems. By converting harmful natural language prompts into formal logical expressions, LogiBreak exploits the distributional gap between alignment data and logic-based inputs, preserving the underlying semantic intent and readability while evading safety constraints. We evaluate LogiBreak on a multilingual jailbreak dataset spanning three languages, demonstrating its effectiveness across various evaluation settings and linguistic contexts.

Robotic foundation models, or generalist robot policies, hold immense potential to enable flexible, general-purpose and dexterous robotic systems. Despite their advancements, our empirical experiments reveal that existing robot policies are prone to learning spurious correlations from pre-training trajectories, adversely affecting their generalization capabilities beyond the training data. To tackle this, we propose a novel Policy Contrastive Decoding (PCD) approach, which redirects the robot policy's focus toward object-relevant visual clues by contrasting action probability distributions derived from original and object-masked visual inputs. As a training-free method, our PCD can be used as a plugin to improve different types of robot policies without needing to finetune or access model weights. We conduct extensive experiments on top of three open-source robot policies, including the autoregressive policy OpenVLA and the diffusion-based policies Octo and $π_0$. The obtained results in both simulation and real-world environments prove PCD's flexibility and effectiveness, e.g., PCD enhances the state-of-the-art policy $π_0$ by 8.9% in the simulation environment and by 108% in the real-world environment. Code and demos are publicly available at: https://koorye.github.io/PCD.

Robot swarms navigating through unknown obstacle environments are an emerging research area that faces challenges. Performing tasks in such environments requires swarms to achieve autonomous localization, perception, decision-making, control, and planning. The limited computational resources of onboard platforms present significant challenges for planning and control. Reactive planners offer low computational demands and high re-planning frequencies but lack predictive capabilities, often resulting in local minima. Multi-step planners can make multi-step predictions to reduce deadlocks, but they require substantial computation, resulting in a lower replanning frequency. This paper proposes a novel homotopic trajectory planning framework for a robot swarm that combines centralized homotopic trajectory planning (optimal virtual tube planning) with distributed control, enabling low-computation, high-frequency replanning, thereby uniting the strengths of multi-step and reactive planners. Based on multi-parametric programming, homotopic optimal trajectories are approximated by affine functions. The resulting approximate solutions have computational complexity $O(n_t)$, where $n_t$ is the number of trajectory parameters. This low complexity makes centralized planning of a large number of optimal trajectories practical and, when combined with distributed control, enables rapid, low-cost replanning.} The effectiveness of the proposed method is validated through several simulations and experiments.

Recent advancements in Multimodal Large Language Models (MLLMs) have demonstrated impressive capabilities in visual-text processing. However, existing static image-text benchmarks are insufficient for evaluating their dynamic perception and interactive reasoning abilities. We introduce Vision-centric Multiple Abilities Game Evaluation (V-MAGE), a novel game-based evaluation framework designed to systematically assess MLLMs' visual reasoning in interactive, continuous-space environments. V-MAGE features five distinct video games comprising over 30 carefully constructed evaluation scenarios. These scenarios are set in free-form, visually complex environments that require models to interpret dynamic game states and make decisions based solely on visual input, thereby closely reflecting the conditions encountered by human players. To ensure robust and interpretable comparisons across models, V-MAGE employs a dynamic ELO-based ranking system that accounts for varying difficulty levels and task diversity. Benchmarking state-of-the-art MLLMs against human baselines reveals that while leading models approach human-level performance in simple tasks, their performance drops significantly in complex scenarios requiring advanced reasoning and task orchestration. This persistent performance gap highlights fundamental limitations in current MLLMs' ability to perform vision-grounded, interactive frame-by-frame control in simulated continuous-time environments. Through extensive analyses, we demonstrate the utility of V-MAGE in uncovering these limitations and providing actionable insights for improving the visual and reasoning capabilities of MLLMs in dynamic, interactive settings. Code is publicly available at https://github.com/CSU-JPG/V-MAGE.

Recent advances in Vision-Language Models (VLMs) have shown promising capabilities in interpreting visualized graph data, offering a new perspective for graph-structured reasoning beyond traditional Graph Neural Networks (GNNs). However, existing studies focus primarily on single-graph reasoning, leaving the critical challenge of multi-graph joint reasoning underexplored. In this work, we introduce the first comprehensive benchmark designed to evaluate and enhance the multi-graph reasoning abilities of VLMs. Our benchmark covers four common graph types-knowledge graphs, flowcharts, mind maps, and route maps-and supports both homogeneous and heterogeneous graph groupings with tasks of increasing complexity. We evaluate several state-of-the-art VLMs under a multi-dimensional scoring framework that assesses graph parsing, reasoning consistency, and instruction-following accuracy. Additionally, we fine-tune multiple open-source models and observe consistent improvements, confirming the effectiveness of our dataset. This work provides a principled step toward advancing multi-graph understanding and reveals new opportunities for cross-modal graph intelligence.

Despite their success, Segment Anything Models (SAMs) experience significant performance drops on severely degraded, low-quality images, limiting their effectiveness in real-world scenarios. To address this, we propose GleSAM, which utilizes Generative Latent space Enhancement to boost robustness on low-quality images, thus enabling generalization across various image qualities. Specifically, we adapt the concept of latent diffusion to SAM-based segmentation frameworks and perform the generative diffusion process in the latent space of SAM to reconstruct high-quality representation, thereby improving segmentation. Additionally, we introduce two techniques to improve compatibility between the pre-trained diffusion model and the segmentation framework. Our method can be applied to pre-trained SAM and SAM2 with only minimal additional learnable parameters, allowing for efficient optimization. We also construct the LQSeg dataset with a greater diversity of degradation types and levels for training and evaluating the model. Extensive experiments demonstrate that GleSAM significantly improves segmentation robustness on complex degradations while maintaining generalization to clear images. Furthermore, GleSAM also performs well on unseen degradations, underscoring the versatility of our approach and dataset.

Cutting-edge robot learning techniques including foundation models and imitation learning from humans all pose huge demands on large-scale and high-quality datasets which constitute one of the bottleneck in the general intelligent robot fields. This paper presents the Kaiwu multimodal dataset to address the missing real-world synchronized multimodal data problems in the sophisticated assembling scenario,especially with dynamics information and its fine-grained labelling. The dataset first provides an integration of human,environment and robot data collection framework with 20 subjects and 30 interaction objects resulting in totally 11,664 instances of integrated actions. For each of the demonstration,hand motions,operation pressures,sounds of the assembling process,multi-view videos, high-precision motion capture information,eye gaze with first-person videos,electromyography signals are all recorded. Fine-grained multi-level annotation based on absolute timestamp,and semantic segmentation labelling are performed. Kaiwu dataset aims to facilitate robot learning,dexterous manipulation,human intention investigation and human-robot collaboration research.

Recent advances in mechanistic interpretability have highlighted the potential of automating interpretability pipelines in analyzing the latent representations within LLMs. While this may enhance our understanding of internal mechanisms, the field lacks standardized evaluation methods for assessing the validity of discovered features. We attempt to bridge this gap by introducing FADE: Feature Alignment to Description Evaluation, a scalable model-agnostic framework for automatically evaluating feature-to-description alignment. FADE evaluates alignment across four key metrics - Clarity, Responsiveness, Purity, and Faithfulness - and systematically quantifies the causes of the misalignment between features and their descriptions. We apply FADE to analyze existing open-source feature descriptions and assess key components of automated interpretability pipelines, aiming to enhance the quality of descriptions. Our findings highlight fundamental challenges in generating feature descriptions, particularly for SAEs compared to MLP neurons, providing insights into the limitations and future directions of automated interpretability. We release FADE as an open-source package at: https://github.com/brunibrun/FADE

Learning from demonstrations is a popular approach to train AI models; however, their vulnerability to adversarial attacks remains underexplored. We present the first systematic study of adversarial attacks, across a range of both classic and recently proposed imitation learning algorithms, including Vanilla Behavior Cloning (Vanilla BC), LSTM-GMM, Implicit Behavior Cloning (IBC), Diffusion Policy (DP), and Vector-Quantized Behavior Transformer (VQ-BET). We study the vulnerability of these methods to both white-box, grey-box and black-box adversarial perturbations. Our experiments reveal that most existing methods are highly vulnerable to these attacks, including black-box transfer attacks that transfer across algorithms. To the best of our knowledge, we are the first to study and compare the vulnerabilities of different popular imitation learning algorithms to both white-box and black-box attacks. Our findings highlight the vulnerabilities of modern imitation learning algorithms, paving the way for future work in addressing such limitations. Videos and code are available at https://sites.google.com/view/uap-attacks-on-bc.

Monte Carlo Tree Search (MCTS) based methods provide promising approaches for generating synthetic data to enhance the self-training of Large Language Model (LLM) based multi-agent systems (MAS). These methods leverage Q-values to estimate individual agent contributions. However, relying solely on Q-values to identify informative data may misalign with the data synthesis objective, as the focus should be on selecting data that best enhances model training. To address this discrepancy, we propose Data Influence-oriented Tree Search (DITS), a novel framework that incorporates influence scores to guide both tree search and data selection. By leveraging influence scores, we effectively identify the most impactful data for system improvement, thereby enhancing model performance. Furthermore, we derive influence score estimation methods tailored for non-differentiable metrics, significantly reducing computational overhead by utilizing inference computations. Extensive experiments on eight multi-agent datasets demonstrate the robustness and effectiveness of the proposed methods. Notably, our findings reveal that allocating more inference resources to estimate influence scores, rather than Q-values, during data synthesis can more effectively and efficiently enhance model training.

Among generative neural models, flow matching techniques stand out for their simple applicability and good scaling properties. Here, velocity fields of curves connecting a simple latent and a target distribution are learned. Then the corresponding ordinary differential equation can be used to sample from a target distribution, starting in samples from the latent one. This paper reviews from a mathematical point of view different techniques to learn the velocity fields of absolutely continuous curves in the Wasserstein geometry. We show how the velocity fields can be characterized and learned via i) transport plans (couplings) between latent and target distributions, ii) Markov kernels and iii) stochastic processes, where the latter two include the coupling approach, but are in general broader. Besides this main goal, we show how flow matching can be used for solving Bayesian inverse problems, where the definition of conditional Wasserstein distances plays a central role. Finally, we briefly address continuous normalizing flows and score matching techniques, which approach the learning of velocity fields of curves from other directions.

Music has always been central to human culture, reflecting and shaping traditions, emotions, and societal changes. Technological advancements have transformed how music is created and consumed, influencing tastes and the music itself. In this study, we use Network Science to analyze musical complexity. Drawing on $\approx20,000$ MIDI files across six macro-genres spanning nearly four centuries, we represent each composition as a weighted directed network to study its structural properties. Our results show that Classical and Jazz compositions have higher complexity and melodic diversity than recently developed genres. However, a temporal analysis reveals a trend toward simplification, with even Classical and Jazz nearing the complexity levels of modern genres. This study highlights how digital tools and streaming platforms shape musical evolution, fostering new genres while driving homogenization and simplicity.

Tensor networks are a tool first employed in the context of many-body quantum physics that now have a wide range of uses across the computational sciences, from numerical methods to machine learning. Methods integrating tensor networks into evolutionary optimization algorithms have appeared in the recent literature. In essence, these methods can be understood as replacing the traditional crossover operation of a genetic algorithm with a tensor network-based generative model. We investigate these methods from the point of view that they are Estimation of Distribution Algorithms (EDAs). We find that optimization performance of these methods is not related to the power of the generative model in a straightforward way. Generative models that are better (in the sense that they better model the distribution from which their training data is drawn) do not necessarily result in better performance of the optimization algorithm they form a part of. This raises the question of how best to incorporate powerful generative models into optimization routines. In light of this we find that adding an explicit mutation operator to the output of the generative model often improves optimization performance.