# Research Papers — 2026-03 · bundle page 231 of 243 > Full-text markdown of papers #11501–11550 of 12123 published in 2026-03, newest first. AI-readable mirrors of the full 12123-paper 2026-03 archive (50 papers per file). Canonical page: https://aipod.dev/papers/archive/2026-03/page/231 [← Previous bundle](https://aipod.dev/papers/archive/2026-03/page/230.md) [Next bundle →](https://aipod.dev/papers/archive/2026-03/page/232.md) ### Symbol-Equivariant Recurrent Reasoning Models Authors: Richard Freinschlag, Timo Bertram, Erich Kobler, Andreas Mayr, Günter Klambauer Published: 2026-03-02 Categories: cs.LG, cs.AI, stat.ML arXiv: https://arxiv.org/abs/2603.02193v1 Reasoning problems such as Sudoku and ARC-AGI remain challenging for neural networks. The structured problem solving architecture family of Recurrent Reasoning Models (RRMs), including Hierarchical Reasoning Model (HRM) and Tiny Recursive Model (TRM), offer a compact alternative to large language models, but currently handle symbol symmetries only implicitly via costly data augmentation. We introduce Symbol-Equivariant Recurrent Reasoning Models (SE-RRMs), which enforce permutation equivariance at the architectural level through symbol-equivariant layers, guaranteeing identical solutions under symbol or color permutations. SE-RRMs outperform prior RRMs on 9x9 Sudoku and generalize from just training on 9x9 to smaller 4x4 and larger 16x16 and 25x25 instances, to which existing RRMs cannot extrapolate. On ARC-AGI-1 and ARC-AGI-2, SE-RRMs achieve competitive performance with substantially less data augmentation and only 2 million parameters, demonstrating that explicitly encoding symmetry improves the robustness and scalability of neural reasoning. Code is available at https://github.com/ml-jku/SE-RRM. --- ### Sketch2Colab: Sketch-Conditioned Multi-Human Animation via Controllable Flow Distillation Authors: Divyanshu Daiya, Aniket Bera Published: 2026-03-02 Categories: cs.CV, cs.AI, cs.GR, cs.HC, cs.LG arXiv: https://arxiv.org/abs/2603.02190v2 We present Sketch2Colab, which turns storyboard-style 2D sketches into coherent, object-aware 3D multi-human motion with fine-grained control over agents, joints, timing, and contacts. Diffusion-based motion generators offer strong realism but often rely on costly guidance for multi-entity control and degrade under strong conditioning. Sketch2Colab instead learns a sketch-conditioned diffusion prior and distills it into a rectified-flow student in latent space for fast, stable sampling. To make motion follow storyboards closely, we guide the student with differentiable objectives that enforce keyframes, paths, contacts, and physical consistency. Collaborative motion naturally involves discrete changes in interaction, such as converging, forming contact, cooperative transport, or disengaging, and a continuous flow alone struggles to sequence these shifts cleanly. We address this with a lightweight continuous-time Markov chain (CTMC) planner that tracks the active interaction regime and modulates the flow to produce clearer, synchronized coordination in human-object-human motion. Experiments on CORE4D and InterHuman show that Sketch2Colab outperforms baselines in constraint adherence and perceptual quality while sampling substantially faster than diffusion-only alternatives. --- ### Multi-Head Low-Rank Attention Authors: Songtao Liu, Hongwu Peng, Zhiwei Zhang, Zhengyu Chen, Yue Guo Published: 2026-03-02 Categories: cs.LG arXiv: https://arxiv.org/abs/2603.02188v1 Long-context inference in large language models is bottlenecked by Key--Value (KV) cache loading during the decoding stage, where the sequential nature of generation requires repeatedly transferring the KV cache from off-chip High-Bandwidth Memory (HBM) to on-chip Static Random-Access Memory (SRAM) at each step. While Multi-Head Latent Attention (MLA) significantly reduces the total KV cache size, it suffers from a sharding bottleneck during distributed decoding via Tensor Parallelism (TP). Since its single latent head cannot be partitioned, each device is forced to redundantly load the complete KV cache for every token, consuming excessive memory traffic and diminishing TP benefits like weight sharding. In this work, we propose Multi-Head Low-Rank Attention (MLRA), which enables partitionable latent states for efficient 4-way TP decoding. Extensive experiments show that MLRA achieves state-of-the-art perplexity and downstream task performance, while also delivering a 2.8$\times$ decoding speedup over MLA. Code is available at https://github.com/SongtaoLiu0823/MLRA. Pretrained weights, along with the training and evaluation data, are available at https://huggingface.co/Soughing/MLRA. --- ### MAC: A Conversion Rate Prediction Benchmark Featuring Labels Under Multiple Attribution Mechanisms Authors: Jinqi Wu, Sishuo Chen, Zhangming Chan, Yong Bai, Lei Zhang, Sheng Chen, Chenghuan Hou, Xiang-Rong Sheng, Han Zhu, Jian Xu, Bo Zheng, Chaoyou Fu Published: 2026-03-02 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2603.02184v2 Multi-attribution learning (MAL), which enhances model performance by learning from conversion labels yielded by multiple attribution mechanisms, has emerged as a promising learning paradigm for conversion rate (CVR) prediction. However, the conversion labels in public CVR datasets are generated by a single attribution mechanism, hindering the development of MAL approaches. To address this data gap, we establish the Multi-Attribution Benchmark (MAC), the first public CVR dataset featuring labels from multiple attribution mechanisms. Besides, to promote reproducible research on MAL, we develop PyMAL, an open-source library covering a wide array of baseline methods. We conduct comprehensive experimental analyses on MAC and reveal three key insights: (1) MAL brings consistent performance gains across different attribution settings, especially for users featuring long conversion paths. (2) The performance growth scales up with objective complexity in most settings; however, when predicting first-click conversion targets, simply adding auxiliary objectives is counterproductive, underscoring the necessity of careful selection of auxiliary objectives. (3) Two architectural design principles are paramount: first, to fully learn the multi-attribution knowledge, and second, to fully leverage this knowledge to serve the main task. Motivated by these findings, we propose Mixture of Asymmetric Experts (MoAE), an effective MAL approach incorporating multi-attribution knowledge learning and main task-centric knowledge utilization. Experiments on MAC show that MoAE substantially surpasses the existing state-of-the-art MAL method. We believe that our benchmark and insights will foster future research in the MAL field. Our MAC benchmark and the PyMAL algorithm library are publicly available at https://github.com/alimama-tech/PyMAL. --- ### MAC: A Conversion Rate Prediction Benchmark Featuring Labels Under Multiple Attribution Mechanisms Authors: Jinqi Wu, Sishuo Chen, Zhangming Chan, Yong Bai, Lei Zhang, Sheng Chen, Chenghuan Hou, Xiang-Rong Sheng, Han Zhu, Jian Xu, Bo Zheng, Chaoyou Fu Published: 2026-03-02 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2603.02184v1 Multi-attribution learning (MAL), which enhances model performance by learning from conversion labels yielded by multiple attribution mechanisms, has emerged as a promising learning paradigm for conversion rate (CVR) prediction. However, the conversion labels in public CVR datasets are generated by a single attribution mechanism, hindering the development of MAL approaches. To address this data gap, we establish the Multi-Attribution Benchmark (MAC), the first public CVR dataset featuring labels from multiple attribution mechanisms. Besides, to promote reproducible research on MAL, we develop PyMAL, an open-source library covering a wide array of baseline methods. We conduct comprehensive experimental analyses on MAC and reveal three key insights: (1) MAL brings consistent performance gains across different attribution settings, especially for users featuring long conversion paths. (2) The performance growth scales up with objective complexity in most settings; however, when predicting first-click conversion targets, simply adding auxiliary objectives is counterproductive, underscoring the necessity of careful selection of auxiliary objectives. (3) Two architectural design principles are paramount: first, to fully learn the multi-attribution knowledge, and second, to fully leverage this knowledge to serve the main task. Motivated by these findings, we propose Mixture of Asymmetric Experts (MoAE), an effective MAL approach incorporating multi-attribution knowledge learning and main task-centric knowledge utilization. Experiments on MAC show that MoAE substantially surpasses the existing state-of-the-art MAL method. We believe that our benchmark and insights will foster future research in the MAL field. Our MAC benchmark and the PyMAL algorithm library are publicly available at https://github.com/alimama-tech/PyMAL. --- ### Leveraging Model Soups to Classify Intangible Cultural Heritage Images from the Mekong Delta Authors: Quoc-Khang Tran, Minh-Thien Nguyen, Nguyen-Khang Pham Published: 2026-03-02 Categories: cs.CV, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2603.02181v1 The classification of Intangible Cultural Heritage (ICH) images in the Mekong Delta poses unique challenges due to limited annotated data, high visual similarity among classes, and domain heterogeneity. In such low-resource settings, conventional deep learning models often suffer from high variance or overfit to spurious correlations, leading to poor generalization. To address these limitations, we propose a robust framework that integrates the hybrid CoAtNet architecture with model soups, a lightweight weight-space ensembling technique that averages checkpoints from a single training trajectory without increasing inference cost. CoAtNet captures both local and global patterns through stage-wise fusion of convolution and self-attention. We apply two ensembling strategies - greedy and uniform soup - to selectively combine diverse checkpoints into a final model. Beyond performance improvements, we analyze the ensembling effect through the lens of bias-variance decomposition. Our findings show that model soups reduces variance by stabilizing predictions across diverse model snapshots, while introducing minimal additional bias. Furthermore, using cross-entropy-based distance metrics and Multidimensional Scaling (MDS), we show that model soups selects geometrically diverse checkpoints, unlike Soft Voting, which blends redundant models centered in output space. Evaluated on the ICH-17 dataset (7,406 images across 17 classes), our approach achieves state-of-the-art results with 72.36% top-1 accuracy and 69.28% macro F1-score, outperforming strong baselines including ResNet-50, DenseNet-121, and ViT. These results underscore that diversity-aware checkpoint averaging provides a principled and efficient way to reduce variance and enhance generalization in culturally rich, data-scarce classification tasks. --- ### Reservoir Subspace Injection for Online ICA under Top-n Whitening Authors: Wenjun Xiao, Yuda Bi, Vince D Calhoun Published: 2026-03-02 Categories: cs.LG, cs.AI, stat.ML arXiv: https://arxiv.org/abs/2603.02178v1 Reservoir expansion can improve online independent component analysis (ICA) under nonlinear mixing, yet top-$n$ whitening may discard injected features. We formalize this bottleneck as \emph{reservoir subspace injection} (RSI): injected features help only if they enter the retained eigenspace without displacing passthrough directions. RSI diagnostics (IER, SSO, $ρ_x$) identify a failure mode in our top-$n$ setting: stronger injection increases IER but crowds out passthrough energy ($ρ_x: 1.00\!\rightarrow\!0.77$), degrading SI-SDR by up to $2.2$\,dB. A guarded RSI controller preserves passthrough retention and recovers mean performance to within $0.1$\,dB of baseline $1/N$ scaling. With passthrough preserved, RE-OICA improves over vanilla online ICA by $+1.7$\,dB under nonlinear mixing and achieves positive SI-SDR$_{\mathrm{sc}}$ on the tested super-Gaussian benchmark ($+0.6$\,dB). --- ### Organizing, Orchestrating, and Benchmarking Agent Skills at Ecosystem Scale Authors: Hao Li, Chunjiang Mu, Jianhao Chen, Siyue Ren, Zhiyao Cui, Yiqun Zhang, Lei Bai, Shuyue Hu Published: 2026-03-02 Categories: cs.CL arXiv: https://arxiv.org/abs/2603.02176v1 The rapid proliferation of Claude agent skills has raised the central question of how to effectively leverage, manage, and scale the agent skill ecosystem. In this paper, we propose AgentSkillOS, the first principled framework for skill selection, orchestration, and ecosystem-level management. AgentSkillOS comprises two stages: (i) Manage Skills, which organizes skills into a capability tree via node-level recursive categorization for efficient discovery; and (ii) Solve Tasks, which retrieves, orchestrates, and executes multiple skills through DAG-based pipelines. To evaluate the agent's ability to invoke skills, we construct a benchmark of 30 artifact-rich tasks across five categories: data computation, document creation, motion video, visual design, and web interaction. We assess the quality of task outputs using LLM-based pairwise evaluation, and the results are aggregated via a Bradley-Terry model to produce unified quality scores. Experiments across three skill ecosystem scales (200 to 200K skills) show that tree-based retrieval effectively approximates oracle skill selection, and that DAG-based orchestration substantially outperforms native flat invocation even when given the identical skill set. Our findings confirm that structured composition is the key to unlocking skill potential. Our GitHub repository is available at:https://github.com/ynulihao/AgentSkillOS. --- ### Kiwi-Edit: Versatile Video Editing via Instruction and Reference Guidance Authors: Yiqi Lin, Guoqiang Liang, Ziyun Zeng, Zechen Bai, Yanzhe Chen, Mike Zheng Shou Published: 2026-03-02 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2603.02175v4 Instruction-based video editing has witnessed rapid progress, yet current methods often struggle with precise visual control, as natural language is inherently limited in describing complex visual nuances. Although reference-guided editing offers a robust solution, its potential is currently bottlenecked by the scarcity of high-quality paired training data. To bridge this gap, we introduce a scalable data generation pipeline that transforms existing video editing pairs into high-fidelity training quadruplets, leveraging image generative models to create synthesized reference scaffolds. Using this pipeline, we construct RefVIE, a large-scale dataset tailored for instruction-reference-following tasks, and establish RefVIE-Bench for comprehensive evaluation. Furthermore, we propose a unified editing architecture, Kiwi-Edit, that synergizes learnable queries and latent visual features for reference semantic guidance. Our model achieves significant gains in instruction following and reference fidelity via a progressive multi-stage training curriculum. Extensive experiments demonstrate that our data and architecture establish a new state-of-the-art in controllable video editing. All datasets, models, and code is released at https://github.com/showlab/Kiwi-Edit. --- ### De-paradox Tree: Breaking Down Simpson's Paradox via A Kernel-Based Partition Algorithm Authors: Xian Teng, Yu-Ru Lin Published: 2026-03-02 Categories: cs.LG arXiv: https://arxiv.org/abs/2603.02174v1 Real-world observational datasets and machine learning have revolutionized data-driven decision-making, yet many models rely on empirical associations that may be misleading due to confounding and subgroup heterogeneity. Simpson's paradox exemplifies this challenge, where aggregated and subgroup-level associations contradict each other, leading to misleading conclusions. Existing methods provide limited support for detecting and interpreting such paradoxical associations, especially for practitioners without deep causal expertise. We introduce De-paradox Tree, an interpretable algorithm designed to uncover hidden subgroup patterns behind paradoxical associations under assumed causal structures involving confounders and effect heterogeneity. It employs novel split criteria and balancing-based procedures to adjust for confounders and homogenize heterogeneous effects through recursive partitioning. Compared to state-of-the-art methods, De-paradox Tree builds simpler, more interpretable trees, selects relevant covariates, and identifies nested opposite effects while ensuring robust estimation of causal effects when causally admissible variables are provided. Our approach addresses the limitations of traditional causal inference and machine learning methods by introducing an interpretable framework that supports non-expert practitioners while explicitly acknowledging causal assumptions and scope limitations, enabling more reliable and informed decision-making in complex observational data environments. --- ### TerraDiT: Point-Conditioned Diffusion Transformer for Satellite Image Synthesis Authors: Srikumar Sastry, Dan Cher, Brian Wei, Aayush Dhakal, Subash Khanal, Dev Gupta, Nathan Jacobs Published: 2026-03-02 Categories: cs.CV arXiv: https://arxiv.org/abs/2603.02172v2 We introduce TerraDiT, a diffusion transformer designed for text-to-satellite image generation with point-based control. Existing controlled satellite image generative models often require pixel-level maps that are time-consuming to acquire, yet semantically limited. To address this limitation, we introduce a novel point-based conditioning framework that controls the generation process through the spatial location of the points and the textual description associated with each point, providing semantically rich control signals. This approach enables flexible, annotation-friendly, and computationally simple inference for satellite image generation. To this end, we introduce an adaptive local attention mechanism that effectively regularizes the attention scores based on the input point queries. We systematically evaluate various domain-specific design choices for training TerraDiT, including the selection of satellite image representation for alignment and geolocation representation for conditioning. Our experiments demonstrate that TerraDiT achieves impressive generation performance, surpassing the state-of-the-art remote sensing generative models. Our models, dataset, and code are available at https://github.com/mvrl/TerraDiT. --- ### SageBwd: A Trainable Low-bit Attention Authors: Jintao Zhang, Marco Chen, Haoxu Wang, Kai Jiang, Ion Stoica, Joseph E. Gonzalez, Jianfei Chen, Jun Zhu Published: 2026-03-02 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2603.02170v1 Low-bit attention, such as SageAttention, has emerged as an effective approach for accelerating model inference, but its applicability to training remains poorly understood. In prior work, we introduced SageBwd, a trainable INT8 attention that quantizes six of seven attention matrix multiplications while preserving fine-tuning performance. However, SageBwd exhibited a persistent performance gap to full-precision attention (FPA) during pre-training. In this work, we investigate why this gap occurs and demonstrate that SageBwd matches full-precision attention during pretraining. Through experiments and theoretical analysis, we reach a few important insights and conclusions: (i) QK-norm is necessary for stable training at large tokens per step, (ii) quantization errors primarily arise from the backward-pass score gradient dS, (iii) reducing tokens per step enables SageBwd to match FPA performance in pre-training, and (iv) K-smoothing remains essential for training stability, while Q-smoothing provides limited benefit during pre-training. --- ### Bridging the gap between Performance and Interpretability: An Explainable Disentangled Multimodal Framework for Cancer Survival Prediction Authors: Aniek Eijpe, Soufyan Lakbir, Melis Erdal Cesur, Sara P. Oliveira, Angelos Chatzimparmpas, Sanne Abeln, Wilson Silva Published: 2026-03-02 Categories: cs.CV arXiv: https://arxiv.org/abs/2603.02162v1 While multimodal survival prediction models are increasingly more accurate, their complexity often reduces interpretability, limiting insight into how different data sources influence predictions. To address this, we introduce DIMAFx, an explainable multimodal framework for cancer survival prediction that produces disentangled, interpretable modality-specific and modality-shared representations from histopathology whole-slide images and transcriptomics data. Across multiple cancer cohorts, DIMAFx achieves state-of-the-art performance and improved representation disentanglement. Leveraging its interpretable design and SHapley Additive exPlanations, DIMAFx systematically reveals key multimodal interactions and the biological information encoded in the disentangled representations. In breast cancer survival prediction, the most predictive features contain modality-shared information, including one capturing solid tumor morphology contextualized primarily by late estrogen response, where higher-grade morphology aligned with pathway upregulation and increased risk, consistent with known breast cancer biology. Key modality-specific features capture microenvironmental signals from interacting adipose and stromal morphologies. These results show that multimodal models can overcome the traditional trade-off between performance and explainability, supporting their application in precision medicine. --- ### Instrumental and Proximal Causal Inference with Gaussian Processes Authors: Yuqi Zhang, Krikamol Muandet, Dino Sejdinovic, Edwin Fong, Siu Lun Chau Published: 2026-03-02 Categories: stat.ML, cs.LG arXiv: https://arxiv.org/abs/2603.02159v2 Instrumental variable (IV) and proximal causal learning (Proxy) methods are central frameworks for causal inference in the presence of unobserved confounding. Despite substantial methodological advances, existing approaches rarely provide reliable epistemic uncertainty (EU) quantification. We address this gap through a Deconditional Gaussian Process (DGP) framework for uncertainty-aware causal learning. Our formulation recovers popular kernel estimators as the posterior mean, ensuring predictive precision, while the posterior variance yields principled and well-calibrated EU. Moreover, the probabilistic structure enables systematic model selection via marginal log-likelihood optimization. Empirical results demonstrate strong predictive performance alongside informative EU quantification, evaluated via empirical coverage frequencies and decision-aware accuracy rejection curves. Together, our approach provides a unified, practical solution for causal inference under unobserved confounding with reliable uncertainty. --- ### How Small Can 6G Reason? Scaling Tiny-to-Small Language Models for AI-Native Networks Authors: Mohamed Amine Ferrag, Abderrahmane Lakas, Merouane Debbah Published: 2026-03-02 Categories: cs.NI, cs.AI arXiv: https://arxiv.org/abs/2603.02156v2 Emerging 6G visions, reflected in ongoing standardization efforts within 3GPP, IETF, ETSI, ITU-T, and the O-RAN Alliance, increasingly characterize networks as AI-native systems in which high-level semantic reasoning layers operate above standardized control and data-plane functions. Although frontier-scale large language models (LLMs) such as Qwen2.5-7B and Olmo-3-7B demonstrate strong reasoning capability, their computational footprint limits deployment in latency-sensitive, edge-native infrastructures. This paper presents a systematic empirical study of the scaling behavior and deployment efficiency of compact language models for network-level semantic reasoning in AI-native 6G systems. Using 6G-Bench, a standardization-aligned benchmark comprising 30 decision-making tasks across five capability domains, we evaluate models ranging from 135M (SmolLM2-135M) to 7B parameters (Qwen2.5-7B), including mid-scale architectures such as Llama-3.2-1B, Granite-1B, and Qwen2.5-3B. Deterministic accuracy (pass@1) increases from 0.224 at 135M to 0.707 at 7B, but scaling gains are highly non-uniform. A pronounced stability transition occurs in the 1 to 1.5B range, where accuracy rises from 0.373 (Llama-3.2-1B) to 0.531 (Qwen2.5-1.5B) and the instability gap Delta_5 contracts from 0.356 to 0.138. Beyond 3B parameters, improvements diminish (+0.064 from 3B to 7B). Through single-query inference profiling and an Edge Score metric that normalizes accuracy by latency and memory footprint, we show that semantic reliability per unit edge resource does not scale monotonically with parameter count. Instead, mid-scale models (approximately 1.5 to 3B) achieve the most favorable balance between deterministic stability and computational efficiency, providing deployment-relevant guidance for AI-native 6G architectures. All scripts and results are publicly available at https://github.com/maferrag/6G-Bench --- ### Near-Optimal Regret for KL-Regularized Multi-Armed Bandits Authors: Kaixuan Ji, Qingyue Zhao, Heyang Zhao, Qiwei Di, Quanquan Gu Published: 2026-03-02 Categories: cs.LG, cs.AI, math.ST, stat.ML arXiv: https://arxiv.org/abs/2603.02155v1 Recent studies have shown that reinforcement learning with KL-regularized objectives can enjoy faster rates of convergence or logarithmic regret, in contrast to the classical $\sqrt{T}$-type regret in the unregularized setting. However, the statistical efficiency of online learning with respect to KL-regularized objectives remains far from completely characterized, even when specialized to multi-armed bandits (MABs). We address this problem for MABs via a sharp analysis of KL-UCB using a novel peeling argument, which yields a $\tilde{O}(ηK\log^2T)$ upper bound: the first high-probability regret bound with linear dependence on $K$. Here, $T$ is the time horizon, $K$ is the number of arms, $η^{-1}$ is the regularization intensity, and $\tilde{O}$ hides all logarithmic factors except those involving $\log T$. The near-tightness of our analysis is certified by the first non-constant lower bound $Ω(ηK \log T)$, which follows from subtle hard-instance constructions and a tailored decomposition of the Bayes prior. Moreover, in the low-regularization regime (i.e., large $η$), we show that the KL-regularized regret for MABs is $η$-independent and scales as $\tildeΘ(\sqrt{KT})$. Overall, our results provide a thorough understanding of KL-regularized MABs across all regimes of $η$ and yield nearly optimal bounds in terms of $K$, $η$, and $T$. --- ### Boltzmann-based Exploration for Robust Decentralized Multi-Agent Planning (Extended Version) Authors: Nhat D. A. Nguyen, Duong D. Nguyen, Gianluca Rizzo, Hung X. Nguyen Published: 2026-03-02 Categories: cs.MA, cs.AI arXiv: https://arxiv.org/abs/2603.02154v2 Decentralized Monte Carlo Tree Search (Dec-MCTS) is widely used for cooperative multi-agent planning but struggles in sparse or skewed reward environments. We introduce Coordinated Boltzmann MCTS (CB-MCTS), which replaces deterministic UCT with a stochastic Boltzmann policy and a decaying entropy bonus for sustained yet focused exploration. While Boltzmann exploration has been studied in single-agent MCTS, applying it in multi-agent systems poses unique challenges. CB-MCTS is the first to address this. We analyze CB-MCTS in the simple-regret setting and show in simulations that it outperforms Dec-MCTS in deceptive scenarios and remains competitive on standard benchmarks, providing a robust solution for multi-agent planning. --- ### Scaling Retrieval Augmented Generation with RAG Fusion: Lessons from an Industry Deployment Authors: Luigi Medrano, Arush Verma, Mukul Chhabra Published: 2026-03-02 Categories: cs.IR, cs.AI, cs.CL arXiv: https://arxiv.org/abs/2603.02153v1 Retrieval-Augmented Generation (RAG) systems commonly adopt retrieval fusion techniques such as multi-query retrieval and reciprocal rank fusion (RRF) to increase document recall, under the assumption that higher recall leads to better answer quality. While these methods show consistent gains in isolated retrieval benchmarks, their effectiveness under realistic production constraints remains underexplored. In this work, we evaluate retrieval fusion in a production-style RAG pipeline operating over an enterprise knowledge base, with fixed retrieval depth, re-ranking budgets, and latency constraints. Across multiple fusion configurations, we find that retrieval fusion does increase raw recall, but these gains are largely neutralized after re-ranking and truncation. In our setting, fusion variants fail to outperform single-query baselines on KB-level Top-$k$ accuracy, with Hit@10 decreasing from $0.51$ to $0.48$ in several configurations. Moreover, fusion introduces additional latency overhead due to query rewriting and larger candidate sets, without corresponding improvements in downstream effectiveness. Our analysis suggests that recall-oriented fusion techniques exhibit diminishing returns once realistic re-ranking limits and context budgets are applied. We conclude that retrieval-level improvements do not reliably translate into end-to-end gains in production RAG systems, and argue for evaluation frameworks that jointly consider retrieval quality, system efficiency, and downstream impact. --- ### Named-Entity Recognition in the Crime Domain (CrimeNER): Case Study and Dataset Authors: Miguel Lopez-Duran, Julian Fierrez, Aythami Morales, Daniel DeAlcala, Gonzalo Mancera, Javier Irigoyen, Ruben Tolosana, Oscar Delgado, Francisco Jurado, Alvaro Ortigosa Published: 2026-03-02 Categories: cs.CL, cs.AI, cs.DB arXiv: https://arxiv.org/abs/2603.02150v2 The extraction of critical information from crime-related documents is a crucial task for law enforcement agencies. The extraction of this information can be interpreted as a Named-Entity Recognition (NER) task. However, there is a considerable lack of adequately annotated data on general real-world crime scenarios. To address this issue, we present CrimeNER, a case study of crime-related NER, and a general crime-related Named-Entity Recognition database (CrimeNER-db), consisting of more than 1.5K annotated documents extracted from public reports of terrorist attacks and the US Department of Justice's press notes. We define 4 coarse types of crime entity and 21 fine-grained entity types. We address the quality of the presented database with experiments using fully supervised finetuned general NER models and zero- and few-shot experiments to address the generalization capabilities. The database is available on GitHub. --- ### 3D Field of Junctions: A Noise-Robust, Training-Free Structural Prior for Volumetric Inverse Problems Authors: Narges Moeini, Namhoon Kim, Justin Romberg, Sara Fridovich-Keil Published: 2026-03-02 Categories: cs.CV, eess.SP arXiv: https://arxiv.org/abs/2603.02149v2 Volume denoising is a foundational problem in computational imaging, as many 3D imaging inverse problems face high levels of measurement noise. Inspired by the strong 2D image denoising properties of Field of Junctions (ICCV 2021), we propose a novel, fully volumetric 3D Field of Junctions (3D FoJ) representation that optimizes a junction of 3D wedges that best explain each 3D patch of a full volume, while encouraging consistency between overlapping patches. In addition to direct volume denoising, we leverage our 3D FoJ representation as a structural prior that: (i) requires no training data, and thus precludes the risk of hallucination, (ii) preserves and enhances sharp edge and corner structures in 3D, even under low signal to noise ratio (SNR), and (iii) can be used as a drop-in denoising representation via projected or proximal gradient descent for any volumetric inverse problem with low SNR. We demonstrate successful volume reconstruction and denoising with 3D FoJ across three diverse 3D imaging tasks with low-SNR measurements: low-dose X-ray computed tomography (CT), cryogenic electron tomography (cryo-ET), and denoising point clouds such as those from lidar in adverse weather. Across these challenging low-SNR volumetric imaging problems, 3D FoJ outperforms the evaluated classical denoisers, untrained neural denoisers, and denoisers trained only on noisy examples. Code is available at https://github.com/voilalab/3D-Field-of-Junctions. --- ### LongRLVR: Long-Context Reinforcement Learning Requires Verifiable Context Rewards Authors: Guanzheng Chen, Michael Qizhe Shieh, Lidong Bing Published: 2026-03-02 Categories: cs.CL arXiv: https://arxiv.org/abs/2603.02146v1 Reinforcement Learning with Verifiable Rewards (RLVR) has significantly advanced the reasoning capabilities of Large Language Models (LLMs) by optimizing them against factual outcomes. However, this paradigm falters in long-context scenarios, as its reliance on internal parametric knowledge is ill-suited for tasks requiring contextual grounding--the ability to find and reason over externally provided information. We identify a key reason for this failure: a reward based solely on the final answer is too sparse to effectively guide the model for identifying relevant evidence. We formally prove that the outcome-only reward leads to significant vanishing gradients for the context grounding process, rendering learning intractable. To overcome this bottleneck, we introduce LongRLVR to augment the sparse answer reward with a dense and verifiable context reward. This auxiliary signal directly incentivizes the model for selecting the correct grounding information, providing a robust learning gradient that solves the underlying optimization challenge. We validate our method on challenging long-context benchmarks using Qwen and LLaMA models. LongRLVR consistently and significantly outperforms the standard RLVR across all models and benchmarks, e.g., boosting a 14B model's scores on RULER-QA from 73.17 to 88.90 and on LongBench v2 from 39.8 to 46.5. Our work demonstrates that explicitly rewarding the grounding process is a critical and effective strategy for unlocking the full reasoning potential of LLMs in long-context applications. Our code is available at https://github.com/real-absolute-AI/LongRLVR. --- ### Machine Learning (ML) library in Linux kernel Authors: Viacheslav Dubeyko Published: 2026-03-02 Categories: cs.LG, cs.OS arXiv: https://arxiv.org/abs/2603.02145v1 Linux kernel is a huge code base with enormous number of subsystems and possible configuration options that results in unmanageable complexity of elaborating an efficient configuration. Machine Learning (ML) is approach/area of learning from data, finding patterns, and making predictions without implementing algorithms by developers that can introduce a self-evolving capability in Linux kernel. However, introduction of ML approaches in Linux kernel is not easy way because there is no direct use of floating-point operations (FPU) in kernel space and, potentially, ML models can be a reason of significant performance degradation in Linux kernel. Paper suggests the ML infrastructure architecture in Linux kernel that can solve the declared problem and introduce of employing ML models in kernel space. Suggested approach of kernel ML library has been implemented as Proof Of Concept (PoC) project with the goal to demonstrate feasibility of the suggestion and to design the interface of interaction the kernel-space ML model proxy and the ML model user-space thread. --- ### When Bigger is Worse: A Practitioner's Guide to Model Selection Under Data Scarcity Authors: Kwame Mbobda-Kuate, Gabriel Kasmi Published: 2026-03-02 Categories: cs.CV, cs.LG arXiv: https://arxiv.org/abs/2603.02142v2 Scaling laws assume larger models trained on more data consistently outperform smaller ones -- an assumption that drives model selection in computer vision but remains untested in resource-constrained Earth observation (EO). We conduct a systematic efficiency analysis across three scaling dimensions: model size, dataset size, and input resolution, on rooftop photovoltaic (PV) detection in Madagascar, yielding 180 training runs across 60 configurations. Optimizing for model efficiency (mAP$_{50}$ per unit of model size), we find a consistent efficiency inversion: YOLO11N achieves the highest efficiency ($22\times$ higher than YOLO11X) with no accuracy penalty: it reaches the second highest absolute mAP$_{50}$ (0.459), outperforming all models except YOLO11S by a margin smaller than run-to-run variance, directly contradicting the scaling prior. Resolution is the dominant resource allocation lever: moving from 416 px to 1280 px at 10 % of the data matches the efficiency gain of collecting the full dataset at low resolution. These findings are robust to the deployment objective: small high-resolution configurations are Pareto-dominant across all 60 experimental setups in the joint accuracy-throughput space, leaving no tradeoff to resolve. In data-scarce EO, the scaling prior does not just fail: it inverts. --- ### Rethinking Camera Choice: An Empirical Study on Fisheye Camera Properties in Robotic Manipulation Authors: Han Xue, Nan Min, Xiaotong Liu, Wendi Chen, Yuan Fang, Jun Lv, Cewu Lu, Chuan Wen Published: 2026-03-02 Categories: cs.RO, cs.CV arXiv: https://arxiv.org/abs/2603.02139v1 The adoption of fisheye cameras in robotic manipulation, driven by their exceptionally wide Field of View (FoV), is rapidly outpacing a systematic understanding of their downstream effects on policy learning. This paper presents the first comprehensive empirical study to bridge this gap, rigorously analyzing the properties of wrist-mounted fisheye cameras for imitation learning. Through extensive experiments in both simulation and the real world, we investigate three critical research questions: spatial localization, scene generalization, and hardware generalization. Our investigation reveals that: (1) The wide FoV significantly enhances spatial localization, but this benefit is critically contingent on the visual complexity of the environment. (2) Fisheye-trained policies, while prone to overfitting in simple scenes, unlock superior scene generalization when trained with sufficient environmental diversity. (3) While naive cross-camera transfer leads to failures, we identify the root cause as scale overfitting and demonstrate that hardware generalization performance can be improved with a simple Random Scale Augmentation (RSA) strategy. Collectively, our findings provide concrete, actionable guidance for the large-scale collection and effective use of fisheye datasets in robotic learning. More results and videos are available on https://robo-fisheye.github.io/ --- ### OmniLottie: Generating Vector Animations via Parameterized Lottie Tokens Authors: Yiying Yang, Wei Cheng, Sijin Chen, Honghao Fu, Xianfang Zeng, Yujun Cai, Gang Yu, Xingjun Ma Published: 2026-03-02 Categories: cs.CV arXiv: https://arxiv.org/abs/2603.02138v1 OmniLottie is a versatile framework that generates high quality vector animations from multi-modal instructions. For flexible motion and visual content control, we focus on Lottie, a light weight JSON formatting for both shapes and animation behaviors representation. However, the raw Lottie JSON files contain extensive invariant structural metadata and formatting tokens, posing significant challenges for learning vector animation generation. Therefore, we introduce a well designed Lottie tokenizer that transforms JSON files into structured sequences of commands and parameters representing shapes, animation functions and control parameters. Such tokenizer enables us to build OmniLottie upon pretrained vision language models to follow multi-modal interleaved instructions and generate high quality vector animations. To further advance research in vector animation generation, we curate MMLottie-2M, a large scale dataset of professionally designed vector animations paired with textual and visual annotations. With extensive experiments, we validate that OmniLottie can produce vivid and semantically aligned vector animations that adhere closely to multi modal human instructions. --- ### NextAds: Towards Next-generation Personalized Video Advertising Authors: Yiyan Xu, Ruoxuan Xia, Wuqiang Zheng, Fengbin Zhu, Wenjie Wang, Fuli Feng Published: 2026-03-02 Categories: cs.IR, cs.CV arXiv: https://arxiv.org/abs/2603.02137v1 With the rapid growth of online video consumption, video advertising has become increasingly dominant in the digital advertising landscape. Yet diverse users and viewing contexts makes one-size-fits-all ad creatives insufficient for consistent effectiveness, underlining the importance of personalization. In practice, most personalized video advertising systems follow a retrieval-based paradigm, selecting the optimal one from a small set of professionally pre-produced creatives for each user. Such static and finite inventories limits both the granularity and the timeliness of personalization, and prevents the creatives from being continuously refined based on online user feedback. Recent advances in generative AI make it possible to move beyond retrieval toward optimizing video creatives in a continuous space at serving time. In this light, we propose NextAds, a generation-based paradigm for next-generation personalized video advertising, and conceptualize NextAds with four core components. To enable comparable research progress, we formulate two representative tasks: personalized creative generation and personalized creative integration, and introduce corresponding lightweight benchmarks. To assess feasibility, we instantiate end-to-end pipelines for both tasks and conduct initial exploratory experiments, demonstrating that GenAI can generate and integrate personalized creatives with encouraging performance. Moreover, we discuss the key challenges and opportunities under this paradigm, aiming to provide actionable insights for both researchers and practitioners and to catalyze progress in personalized video advertising. --- ### OnlineX: Unified Online 3D Reconstruction and Understanding with Active-to-Stable State Evolution Authors: Chong Xia, Fangfu Liu, Yule Wang, Yize Pang, Yueqi Duan Published: 2026-03-02 Categories: cs.CV arXiv: https://arxiv.org/abs/2603.02134v2 Recent advances in generalizable 3D Gaussian Splatting (3DGS) have enabled rapid 3D scene reconstruction within seconds, eliminating the need for per-scene optimization. However, existing methods primarily follow an offline reconstruction paradigm, lacking the capacity for continuous reconstruction, which limits their applicability to online scenarios such as robotics and VR/AR. In this paper, we introduce OnlineX, a feed-forward framework that reconstructs both 3D visual appearance and language fields in an online manner using only streaming images. A key challenge in online formulation is the cumulative drift issue, which is rooted in the fundamental conflict between two opposing roles of the memory state: an active role that constantly refreshes to capture high-frequency local geometry, and a stable role that conservatively accumulates and preserves the long-term global structure. To address this, we introduce a decoupled active-to-stable state evolution paradigm. Our framework decouples the memory state into a dedicated active state and a persistent stable state, and then cohesively fuses the information from the former into the latter to achieve both fidelity and stability. Moreover, we jointly model visual appearance and language fields and incorporate an implicit Gaussian fusion module to enhance reconstruction quality. Experiments on mainstream datasets demonstrate that our method consistently outperforms prior work in novel view synthesis and semantic understanding, showcasing robust performance across input sequences of varying lengths with real-time inference speed. --- ### SimRecon: SimReady Compositional Scene Reconstruction from Real Videos Authors: Chong Xia, Kai Zhu, Zizhuo Wang, Fangfu Liu, Zhizheng Zhang, Yueqi Duan Published: 2026-03-02 Categories: cs.CV arXiv: https://arxiv.org/abs/2603.02133v2 Compositional scene reconstruction seeks to create object-centric representations rather than holistic scenes from real-world videos, which is natively applicable for simulation and interaction. Conventional compositional reconstruction approaches primarily emphasize on visual appearance and show limited generalization ability to real-world scenarios. In this paper, we propose SimRecon, a framework that realizes a "Perception-Generation-Simulation" pipeline towards cluttered scene reconstruction, which first conducts scene-level semantic reconstruction from video input, then performs single-object generation, and finally assembles these assets in the simulator. However, naively combining these three stages leads to visual infidelity of generated assets and physical implausibility of the final scene, a problem particularly severe for complex scenes. Thus, we further propose two bridging modules between the three stages to address this problem. To be specific, for the transition from Perception to Generation, critical for visual fidelity, we introduce Active Viewpoint Optimization, which actively searches in 3D space to acquire optimal projected images as conditions for single-object completion. Moreover, for the transition from Generation to Simulation, essential for physical plausibility, we propose a Scene Graph Synthesizer, which guides the construction from scratch in 3D simulators, mirroring the native, constructive principle of the real world. Extensive experiments on the ScanNet dataset validate our method's superior performance over previous state-of-the-art approaches. --- ### Stereo-Inertial Poser: Towards Metric-Accurate Shape-Aware Motion Capture Using Sparse IMUs and a Single Stereo Camera Authors: Tutian Tang, Xingyu Ji, Yutong Li, MingHao Liu, Wenqiang Xu, Cewu Lu Published: 2026-03-02 Categories: cs.CV arXiv: https://arxiv.org/abs/2603.02130v1 Recent advancements in visual-inertial motion capture systems have demonstrated the potential of combining monocular cameras with sparse inertial measurement units (IMUs) as cost-effective solutions, which effectively mitigate occlusion and drift issues inherent in single-modality systems. However, they are still limited by metric inaccuracies in global translations stemming from monocular depth ambiguity, and shape-agnostic local motion estimations that ignore anthropometric variations. We present Stereo-Inertial Poser, a real-time motion capture system that leverages a single stereo camera and six IMUs to estimate metric-accurate and shape-aware 3D human motion. By replacing the monocular RGB with stereo vision, our system resolves depth ambiguity through calibrated baseline geometry, enabling direct 3D keypoint extraction and body shape parameter estimation. IMU data and visual cues are fused for predicting drift-compensated joint positions and root movements, while a novel shape-aware fusion module dynamically harmonizes anthropometry variations with global translations. Our end-to-end pipeline achieves over 200 FPS without optimization-based post-processing, enabling real-time deployment. Quantitative evaluations across various datasets demonstrate state-of-the-art performance. Qualitative results show our method produces drift-free global translation under a long recording time and reduces foot-skating effects. --- ### LiftAvatar: Kinematic-Space Completion for Expression-Controlled 3D Gaussian Avatar Animation Authors: Hualiang Wei, Shunran Jia, Jialun Liu, Wenhui Li Published: 2026-03-02 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2603.02129v1 We present LiftAvatar, a new paradigm that completes sparse monocular observations in kinematic space (e.g., facial expressions and head pose) and uses the completed signals to drive high-fidelity avatar animation. LiftAvatar is a fine-grained, expression-controllable large-scale video diffusion Transformer that synthesizes high-quality, temporally coherent expression sequences conditioned on single or multiple reference images. The key idea is to lift incomplete input data into a richer kinematic representation, thereby strengthening both reconstruction and animation in downstream 3D avatar pipelines. To this end, we introduce (i) a multi-granularity expression control scheme that combines shading maps with expression coefficients for precise and stable driving, and (ii) a multi-reference conditioning mechanism that aggregates complementary cues from multiple frames, enabling strong 3D consistency and controllability. As a plug-and-play enhancer, LiftAvatar directly addresses the limited expressiveness and reconstruction artifacts of 3D Gaussian Splatting-based avatars caused by sparse kinematic cues in everyday monocular videos. By expanding incomplete observations into diverse pose-expression variations, LiftAvatar also enables effective prior distillation from large-scale video generative models into 3D pipelines, leading to substantial gains. Extensive experiments show that LiftAvatar consistently boosts animation quality and quantitative metrics of state-of-the-art 3D avatar methods, especially under extreme, unseen expressions. --- ### LLMs as Strategic Actors: Behavioral Alignment, Risk Calibration, and Argumentation Framing in Geopolitical Simulations Authors: Veronika Solopova, Viktoria Skorik, Maksym Tereshchenko, Alina Haidun, Ostap Vykhopen Published: 2026-03-02 Categories: cs.CL, cs.AI, cs.CY arXiv: https://arxiv.org/abs/2603.02128v1 Large language models (LLMs) are increasingly proposed as agents in strategic decision environments, yet their behavior in structured geopolitical simulations remains under-researched. We evaluate six popular state-of-the-art LLMs alongside results from human results across four real-world crisis simulation scenarios, requiring models to select predefined actions and justify their decisions across multiple rounds. We compare models to humans in action alignment, risk calibration through chosen actions' severity, and argumentative framing grounded in international relations theory. Results show that models approximate human decision patterns in base simulation rounds but diverge over time, displaying distinct behavioural profiles and strategy updates. LLM explanations for chosen actions across all models exhibit a strong normative-cooperative framing centered on stability, coordination, and risk mitigation, with limited adversarial reasoning. --- ### A 3D mesh convolution-based autoencoder for geometry compression Authors: Germain Bregeon, Marius Preda, Radu Ispas, Titus Zaharia Published: 2026-03-02 Categories: cs.CV arXiv: https://arxiv.org/abs/2603.02125v1 In this paper, we introduce a novel 3D mesh convolution-based autoencoder for geometry compression, able to deal with irregular mesh data without requiring neither preprocessing nor manifold/watertightness conditions. The proposed approach extracts meaningful latent representations by learning features directly from the mesh faces, while preserving connectivity through dedicated pooling and unpooling operations. The encoder compresses the input mesh into a compact base mesh space, which ensures that the latent space remains comparable. The decoder reconstructs the original connectivity and restores the compressed geometry to its full resolution. Extensive experiments on multi-class datasets demonstrate that our method outperforms state-of-the-art approaches in both 3D mesh geometry reconstruction and latent space classification tasks. Code available at: github.com/germainGB/MeshConv3D --- ### Nano-EmoX: Unifying Multimodal Emotional Intelligence from Perception to Empathy Authors: Jiahao Huang, Fengyan Lin, Xuechao Yang, Chen Feng, Kexin Zhu, Xu Yang, Zhide Chen Published: 2026-03-02 Categories: cs.AI, cs.CV arXiv: https://arxiv.org/abs/2603.02123v3 The development of affective multimodal language models (MLMs) has long been constrained by a gap between low-level perception and high-level interaction, leading to fragmented affective capabilities and limited generalization. To bridge this gap, we propose a cognitively inspired three-level hierarchy that organizes affective tasks according to their cognitive depth-perception, understanding, and interaction-and provides a unified conceptual foundation for advancing affective modeling. Guided by this hierarchy, we introduce Nano-EmoX, a small-scale multitask MLM, and P2E (Perception-to-Empathy), a curriculum-based training framework. Nano-EmoX integrates a suite of omni-modal encoders, including an enhanced facial encoder and a fusion encoder, to capture key multimodal affective cues and improve cross-task transferability. The outputs are projected into a unified language space via heterogeneous adapters, empowering a lightweight language model to tackle diverse affective tasks. Concurrently, P2E progressively cultivates emotional intelligence by aligning rapid perception with chain-of-thought-driven empathy. To the best of our knowledge, Nano-EmoX is the first compact MLM (2.2B) to unify six core affective tasks across all three hierarchy levels, achieving state-of-the-art or highly competitive performance across multiple benchmarks, demonstrating excellent efficiency and generalization. The code is available at https://github.com/waHAHJIAHAO/Nano-EmoX. --- ### Pencil Puzzle Bench: A Benchmark for Multi-Step Verifiable Reasoning Authors: Justin Waugh Published: 2026-03-02 Categories: cs.AI, cs.GT, cs.LG arXiv: https://arxiv.org/abs/2603.02119v1 We introduce Pencil Puzzle Bench, a framework for evaluating large language model reasoning through pencil puzzles, a family of constraint-satisfaction problems closely related to NP-complete problems, with deterministic, step-level verification. From a database of 62,231 puzzles across 94 varieties with verified unique solutions, we select a benchmark of 300 puzzles spanning 20 varieties and evaluate 51 models from 11 providers in two modes: direct ask (single-shot) and agentic (multi-turn with iterative verification). A key differentiator of our benchmark is that every intermediate board state can be checked against variety-specific constraints, localizing errors to the exact rule violated, providing the infrastructure for dense, per-move reward signals for process supervision and reinforcement learning. Our evaluation reveals two distinct axes of capability: (1) reasoning effort scaling, where GPT-5.2 improves 81x from no reasoning to maximum effort; and (2) agentic iteration, where Claude Opus 4.6 rises from 0.3% to 30.0% through iterative checking, while GPT-5.2@xhigh improves from 20.2% to 56.0%. Agentic attempts span a median of 29 turns over 17 minutes, with the longest exceeding 1,221 turns and 14.3 hours - a demanding test of long-context utilization, not just reasoning. --- ### Robometer: Scaling General-Purpose Robotic Reward Models via Trajectory Comparisons Authors: Anthony Liang, Yigit Korkmaz, Jiahui Zhang, Minyoung Hwang, Abrar Anwar, Sidhant Kaushik, Aditya Shah, Alex S. Huang, Luke Zettlemoyer, Dieter Fox, Yu Xiang, Anqi Li et al. Published: 2026-03-02 Categories: cs.RO, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2603.02115v2 General-purpose robot reward models are typically trained to predict absolute task progress from expert demonstrations, providing only local, frame-level supervision. While effective for expert demonstrations, this paradigm scales poorly to large-scale robotics datasets where failed and suboptimal trajectories are abundant and assigning dense progress labels is ambiguous. We introduce Robometer, a scalable reward modeling framework that combines intra-trajectory progress supervision with inter-trajectory preference supervision. Robometer is trained with a dual objective: a frame-level progress loss that anchors reward magnitude on expert data, and a trajectory-comparison preference loss that imposes global ordering constraints across trajectories of the same task, enabling effective learning from both real and augmented failed trajectories. To support this formulation at scale, we curate RBM-1M, a reward-learning dataset comprising over one million trajectories spanning diverse robot embodiments and tasks, including substantial suboptimal and failure data. Across benchmarks and real-world evaluations, Robometer learns more generalizable reward functions than prior methods and improves robot learning performance across a diverse set of downstream applications. Code, model weights, and videos at https://robometer.github.io/. --- ### Recursive Models for Long-Horizon Reasoning Authors: Chenxiao Yang, Nathan Srebro, Zhiyuan Li Published: 2026-03-02 Categories: cs.LG, cs.CL arXiv: https://arxiv.org/abs/2603.02112v2 Modern language models reason within bounded context, an inherent constraint that poses a fundamental barrier to long-horizon reasoning. We identify recursion as a core principle for overcoming this barrier, and propose recursive models as a minimal realization, where the model can recursively invoke itself to solve subtasks in isolated contexts. We prove that any computable problem admits a recursive decomposition of reasoning in which each subtask requires only exponentially smaller active context than standard autoregressive models; this strictly surpasses any context management approach confined to a single sequence, such as summarization. We further generalize our framework to modern agentic systems with arbitrary context processing and control flows, and prove that recursive models can achieve optimal power within this broader class. Experimentally, we test two settings: fine-tuning a pretrained base model for recursive SAT solving, and training a small model from scratch on Go traces generated by exact game-tree search. Both show improved long-horizon accuracy with small active contexts. --- ### Orchestrating Multimodal DNN Workloads in Wireless Neural Processing Authors: Sai Xu, Kai-Kit Wong, Yanan Du, Hyundong Shin Published: 2026-03-02 Categories: eess.SP, cs.LG arXiv: https://arxiv.org/abs/2603.02109v1 In edge inference, wireless resource allocation and accelerator-level deep neural network (DNN) scheduling have yet to be co-optimized in an end-to-end manner. The lack of coordination between wireless transmission and accelerator-level DNN execution prevents efficient overlap, leading to higher end-to-end inference latency. To address this issue, this paper investigates multimodal DNN workload orchestration in wireless neural processing (WNP), a paradigm that integrates wireless transmission and multi-core accelerator execution into a unified end-to-end pipeline. First, we develop a unified communication-computation model for multimodal DNN execution and formulate the corresponding optimization problem. Second, we propose O-WiN, a framework that orchestrates DNN workloads in WNP through two tightly coupled stages: simulation-based optimization and runtime execution. Third, we develop two algorithms, RTFS and PACS. RTFS schedules communication and computation sequentially, whereas PACS interleaves them to enable pipeline parallelism by overlapping wireless data transfer with accelerator-level DNN execution. Simulation results demonstrate that PACS significantly outperforms RTFS under high modality heterogeneity by better masking wireless latency through communication-computation overlap, thereby highlighting the effectiveness of communication-computation pipelining in accelerating multimodal DNN execution in WNP. --- ### Stochastic Multi-Armed Bandits with Limited Control Variates Authors: Arun Verma, Manjesh Kumar Hanawal, Arun Rajkumar Published: 2026-03-02 Categories: cs.LG arXiv: https://arxiv.org/abs/2603.02100v1 Motivated by wireless networks where interference or channel state estimates provide partial insight into throughput, we study a variant of the classical stochastic multi-armed bandit problem in which the learner has limited access to auxiliary information. Recent work has shown that such auxiliary information, when available as control variates, can be used to get tighter confidence bounds, leading to lower regret. However, existing works assume that control variates are available in every round, which may not be realistic in several real-life scenarios. To address this, we propose UCB-LCV, an upper confidence bound (UCB) based algorithm that effectively combines the estimators obtained from rewards and control variates. When there is no control variate, UCB-LCV leads to a novel algorithm that we call UCB-NORMAL, outperforming its existing algorithms for the standard MAB setting with normally distributed rewards. Finally, we discuss variants of the proposed UCB-LCV that apply to general distributions and experimentally demonstrate that UCB-LCV outperforms existing bandit algorithms. --- ### Recursive Think-Answer Process for LLMs and VLMs Authors: Byung-Kwan Lee, Youngchae Chee, Yong Man Ro Published: 2026-03-02 Categories: cs.CL arXiv: https://arxiv.org/abs/2603.02099v2 Think-Answer reasoners such as DeepSeek-R1 have made notable progress by leveraging interpretable internal reasoning. However, despite the frequent presence of self-reflective cues like "Oops!", they remain vulnerable to output errors during single-pass inference. To address this limitation, we propose an efficient Recursive Think-Answer Process (R-TAP) that enables models to engage in iterative reasoning cycles and generate more accurate answers, going beyond conventional single-pass approaches. Central to this approach is a confidence generator that evaluates the certainty of model responses and guides subsequent improvements. By incorporating two complementary rewards-Recursively Confidence Increase Reward and Final Answer Confidence Reward-we show that R-TAP-enhanced models consistently outperform conventional single-pass methods for both large language models (LLMs) and vision-language models (VLMs). Moreover, by analyzing the frequency of "Oops"-like expressions in model responses, we find that R-TAP-applied models exhibit significantly fewer self-reflective patterns, resulting in more stable and faster inference-time reasoning. We hope R-TAP pave the way evolving into efficient and elaborated methods to refine the reasoning processes of future AI. --- ### Efficient and High-Fidelity Omni Modality Retrieval Authors: Chuong Huynh, Manh Luong, Abhinav Shrivastava Published: 2026-03-02 Categories: cs.IR, cs.CL, cs.CV arXiv: https://arxiv.org/abs/2603.02098v2 Multimodal retrieval is the task of aggregating information from queries across heterogeneous modalities to retrieve desired targets. State-of-the-art multimodal retrieval models can understand complex queries, yet they are typically limited to two modalities: text and vision. This limitation impedes the development of universal retrieval systems capable of comprehending queries that combine more than two modalities. To advance toward this goal, we present OmniRet, the first retrieval model capable of handling complex, composed queries spanning three key modalities: text, vision, and audio. Our OmniRet model addresses two critical challenges for universal retrieval: computational efficiency and representation fidelity. First, feeding massive token sequences from modality-specific encoders to Large Language Models (LLMs) is computationally inefficient. We therefore introduce an attention-based resampling mechanism to generate compact, fixed-size representations from these sequences. Second, compressing rich omni-modal data into a single embedding vector inevitably causes information loss and discards fine-grained details. We propose Attention Sliced Wasserstein Pooling to preserve these fine-grained details, leading to improved omni-modal representations. OmniRet is trained on an aggregation of approximately 6 million query-target pairs spanning 30 datasets. We benchmark our model on 13 retrieval tasks and a MMEBv2 subset. Our model demonstrates significant improvements on composed query, audio and video retrieval tasks, while achieving on-par performance with state-of-the-art models on others. Furthermore, we curate a new Audio-Centric Multimodal Benchmark (ACM). This new benchmark introduces two critical, previously missing tasks-composed audio retrieval and audio-visual retrieval to more comprehensively evaluate a model's omni-modal embedding capacity. --- ### ClinConsensus: A Physician-Calibrated Benchmark for Evaluating Clinical Rubric Coverage in Chinese Medical LLMs Authors: Xiang Zheng, Han Li, Wenjie Luo, Weiqi Zhai, Yiyuan Li, Chuanmiao Yan, Xue Yang, Kailuan Wu, Ruyi Xu, Tianyun Lu, Tianyi Tang, Yubo Ma et al. Published: 2026-03-02 Categories: cs.CL arXiv: https://arxiv.org/abs/2603.02097v6 Open-ended medical LLM evaluation remains weakly grounded in physician-calibrated coverage of clinically relevant response criteria, especially in localized clinical settings. We introduce \textsc{ClinConsensus}, a Chinese medical benchmark of 2{,}500 expert-curated cases spanning 36 specialties, 12 task themes, multiple difficulty levels, and lay-facing versus professional-facing settings. Each case is paired with 30 case-specific binary rubric criteria. To evaluate whether responses satisfy enough physician-authored criteria, we propose \emph{Clinician-Anchored Coverage Score} (CACS), a physician-calibrated threshold metric instantiated at \(k=10\), and develop a dual-judge framework combining a GPT-5.1 grader with a physician-supervised Qwen3-8B judge. Evaluating 11 frontier LLMs, we find a persistent coverage gap: Rubric Accuracy ranges from 39.6\% to 52.1\%, whereas CACS@10 ranges from 17.8\% to 32.9\%, leaving a 19.2--21.9 point gap across models. Stratified analyses further reveal substantial variation across reasoning, evidence use, structured extraction, medication instructions, follow-up, and dialogue register. These results suggest that medical LLM evaluation should measure thresholded, rubric-grounded clinical coverage rather than average partial correctness. --- ### ClinConsensus: A Physician-Calibrated Benchmark for Evaluating Clinical Rubric Coverage in Chinese Medical LLMs Authors: Xiang Zheng, Han Li, Wenjie Luo, Weiqi Zhai, Yiyuan Li, Chuanmiao Yan, Xue Yang, Kailuan Wu, Ruyi Xu, Tianyun Lu, Tianyi Tang, Yubo Ma et al. Published: 2026-03-02 Categories: cs.CL arXiv: https://arxiv.org/abs/2603.02097v5 Open-ended medical LLM evaluation remains weakly grounded in physician-calibrated coverage of clinically relevant response criteria, especially in localized clinical settings. We introduce \textsc{ClinConsensus}, a Chinese medical benchmark of 2{,}500 expert-curated cases spanning 36 specialties, 12 task themes, multiple difficulty levels, and lay-facing versus professional-facing settings. Each case is paired with 30 case-specific binary rubric criteria. To evaluate whether responses satisfy enough physician-authored criteria, we propose \emph{Clinician-Anchored Coverage Score} (CACS), a physician-calibrated threshold metric instantiated at \(k=10\), and develop a dual-judge framework combining a GPT-5.1 grader with a physician-supervised Qwen3-8B judge. Evaluating 11 frontier LLMs, we find a persistent coverage gap: Rubric Accuracy ranges from 39.6\% to 52.1\%, whereas CACS@10 ranges from 17.8\% to 32.9\%, leaving a 19.2--21.9 point gap across models. Stratified analyses further reveal substantial variation across reasoning, evidence use, structured extraction, medication instructions, follow-up, and dialogue register. These results suggest that medical LLM evaluation should measure thresholded, rubric-grounded clinical coverage rather than average partial correctness. --- ### FluxMem: Adaptive Hierarchical Memory for Streaming Video Understanding Authors: Yiweng Xie, Bo He, Junke Wang, Xiangyu Zheng, Ziyi Ye, Zuxuan Wu Published: 2026-03-02 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2603.02096v1 This paper presents FluxMem, a training-free framework for efficient streaming video understanding. FluxMem adaptively compresses redundant visual memory through a hierarchical, two-stage design: (1) a Temporal Adjacency Selection (TAS) module removes redundant visual tokens across adjacent frames, and (2) a Spatial Domain Consolidation (SDC) module further merges spatially repetitive regions within each frame into compact representations. To adapt effectively to dynamic scenes, we introduce a self-adaptive token compression mechanism in both TAS and SDC, which automatically determines the compression rate based on intrinsic scene statistics rather than manual tuning. Extensive experiments demonstrate that FluxMem achieves new state-of-the-art results on existing online video benchmarks, reaching 76.4 on StreamingBench and 67.2 on OVO-Bench under real-time settings, while reducing latency by 69.9% and peak GPU memory by 34.5% on OVO-Bench. Furthermore, it maintains strong offline performance, achieving 73.1 on MLVU while using 65% fewer visual tokens. --- ### On the Rate of Convergence of GD in Non-linear Neural Networks: An Adversarial Robustness Perspective Authors: Guy Smorodinsky, Sveta Gimpleson, Itay Safran Published: 2026-03-02 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2603.02095v1 We study the convergence dynamics of Gradient Descent (GD) in a minimal binary classification setting, consisting of a two-neuron ReLU network and two training instances. We prove that even under these strong simplifying assumptions, while GD successfully converges to an optimal robustness margin, effectively maximizing the distance between the decision boundary and the training points, this convergence occurs at a prohibitively slow rate, scaling strictly as $Θ(1/\ln(t))$. To the best of our knowledge, this establishes the first explicit lower bound on the convergence rate of the robustness margin in a non-linear model. Through empirical simulations, we further demonstrate that this inherent failure mode is pervasive, exhibiting the exact same tight convergence rate across multiple natural network initializations. Our theoretical guarantees are derived via a rigorous analysis of the GD trajectories across the distinct activation patterns of the model. Specifically, we develop tight control over the system's dynamics to bound the trajectory of the decision boundary, overcoming the primary technical challenge introduced by the non-linear nature of the architecture. --- ### Adam Converges Without Any Modification On Update Rules Authors: Yushun Zhang, Bingran Li, Congliang Chen, Zhi-Quan Luo, Ruoyu Sun Published: 2026-03-02 Categories: cs.LG, math.OC arXiv: https://arxiv.org/abs/2603.02092v1 Adam is the default algorithm for training neural networks, including large language models (LLMs). However, \citet{reddi2019convergence} provided an example that Adam diverges, raising concerns for its deployment in AI model training. We identify a key mismatch between the divergence example and practice: \citet{reddi2019convergence} pick the problem after picking the hyperparameters of Adam, i.e., $(β_1,β_2)$; while practical applications often fix the problem first and then tune $(β_1,β_2)$. In this work, we prove that Adam converges with proper problem-dependent hyperparameters. First, we prove that Adam converges when $β_2$ is large and $β_1 < \sqrt{β_2}$. Second, when $β_2$ is small, we point out a region of $(β_1,β_2)$ combinations where Adam can diverge to infinity. Our results indicate a phase transition for Adam from divergence to convergence when changing the $(β_1, β_2)$ combination. To our knowledge, this is the first phase transition in $(β_1,β_2)$ 2D-plane reported in the literature, providing rigorous theoretical guarantees for Adam optimizer. We further point out that the critical boundary $(β_1^*, β_2^*)$ is problem-dependent, and particularly, dependent on batch size. This provides suggestions on how to tune $β_1$ and $β_2$: when Adam does not work well, we suggest tuning up $β_2$ inversely with batch size to surpass the threshold $β_2^*$, and then trying $β_1< \sqrt{β_2}$. Our suggestions are supported by reports from several empirical studies, which observe improved LLM training performance when applying them. --- ### Learning from Synthetic Data Improves Multi-hop Reasoning Authors: Anmol Kabra, Yilun Yin, Albert Gong, Kamilė Stankevičiūtė, Dongyoung Go, Johann Lee, Katie Z. Luo, Carla P. Gomes, Kilian Q. Weinberger Published: 2026-03-02 Categories: cs.LG, cs.AI, cs.CL arXiv: https://arxiv.org/abs/2603.02091v1 Reinforcement Learning (RL) has been shown to significantly boost reasoning capabilities of large language models (LLMs) in math, coding, and multi-hop reasoning tasks. However, RL fine-tuning requires abundant high-quality verifiable data, often sourced from human annotations, generated from frontier LLMs, or scored by LLM-based verifiers. All three have considerable limitations: human-annotated datasets are small and expensive to curate, LLM-generated data is hallucination-prone and costly, and LLM-based verifiers are inaccurate and slow. In this work, we investigate a cheaper alternative: RL fine-tuning on rule-generated synthetic data for multi-hop reasoning tasks. We discover that LLMs fine-tuned on synthetic data perform significantly better on popular real-world question-answering benchmarks, despite the synthetic data containing only fictional knowledge. On stratifying performance by question difficulty, we find that synthetic data teaches LLMs to compose knowledge -- a fundamental and generalizable reasoning skill. Our work highlights rule-generated synthetic reasoning data as a free and scalable resource to improve LLM reasoning capabilities. --- ### A Detection-Gated Pipeline for Robust Glottal Area Waveform Extraction and Clinical Pathology Assessment Authors: Harikrishnan Unnikrishnan, Rita Patel Published: 2026-03-02 Categories: cs.CV, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2603.02087v3 We present a fully automated, two-stage modular glottal area segmentation framework for high-speed videoendoscopy (HSV) designed for accuracy, generalizability, and real-time playback. Our detection-gated pipeline combines a YOLOv8n glottis localizer with a U-Net segmenter; the localizer defines a tight crop to ensure a consistent field of view and gates the output to reduce spurious segmentations during glottal closure. The models were trained on the GIRAFE (N=600) and BAGLS (N=55,750) datasets. Cross-dataset portability was evaluated by benchmarking GIRAFE-trained models on the BAGLS test set without fine-tuning. In these evaluations, the pipeline achieved a Dice Similarity Coefficient (DSC) of 0.745 (87% of the in-domain ceiling). On in-distribution test sets, the system achieved DSCs of 0.81 (GIRAFE) and 0.856 (BAGLS), outperforming or competing with state-of-the-art methods. An exploratory clinical study of 40 subjects demonstrated that the glottal area Coefficient of Variation (CV distinguished healthy from pathological function (p=0.006). The system processes ~35 frames per second on commodity hardware, enabling interactive clinical review. This design supports uniform extraction of laryngeal kinematic measures across varying acquisition settings. Code, weights, and software are available at https://github.com/hari-krishnan/openglottal. --- ### Modeling Grammatical Hypothesis Testing in Young Learners: A Sequence-Based Learning Analytics Study of Morphosyntactic Reasoning in an Interactive Game Authors: Thierry Geoffre, Trystan Geoffre Published: 2026-03-02 Categories: cs.CL arXiv: https://arxiv.org/abs/2603.02084v1 This study investigates grammatical reasoning in primary school learners through a sequence-based learning analytics approach, leveraging fine-grained action sequences from an interactive game targeting morphosyntactic agreement in French. Unlike traditional assessments that rely on final answers, we treat each slider movement as a hypothesis-testing action, capturing real-time cognitive strategies during sentence construction. Analyzing 597 gameplay sessions (9,783 actions) from 100 students aged 8-11 in authentic classroom settings, we introduce Hamming distance to quantify proximity to valid grammatical solutions and examine convergence patterns across exercises with varying levels of difficulty. Results reveal that determiners and verbs are key sites of difficulty, with action sequences deviating from left-to-right usual treatment. This suggests learners often fix the verb first and adjust preceding elements. Exercises with fewer solutions exhibit slower and more erratic convergence, while changes in the closest valid solution indicate dynamic hypothesis revision. Our findings demonstrate how sequence-based analytics can uncover hidden dimensions of linguistic reasoning, offering a foundation for real-time scaffolding and teacher-facing tools in linguistically diverse classrooms. --- ### $π$-StepNFT: Wider Space Needs Finer Steps in Online RL for Flow-based VLAs Authors: Siting Wang, Xiaofeng Wang, Zheng Zhu, Minnan Pei, Xinyu Cui, Cheng Deng, Jian Zhao, Guan Huang, Haifeng Zhang, Jun Wang Published: 2026-03-02 Categories: cs.RO, cs.CV arXiv: https://arxiv.org/abs/2603.02083v2 Flow-based vision-language-action (VLA) models excel in embodied control but suffer from intractable likelihoods during multi-step sampling, hindering online reinforcement learning. We propose \textbf{\textit{$\boldsymbolπ$-StepNFT}} (Step-wise Negative-aware Fine-Tuning), a critic-and-likelihood-free framework that requires only a single forward pass per optimization step and eliminates auxiliary value networks. We identify that wider exploration spaces necessitate finer-grained, step-wise guidance for alignment. Empirically, $π$-StepNFT unlocks latent potential on LIBERO with competitive few-shot robustness. Moreover, it achieves superior generalization on ManiSkill, outperforming value-based baselines in OOD scenarios by preventing overfitting to multimodal features. This property offers a scalable solution promising for complex real-world applications. --- ### What Exactly do Children Receive in Language Acquisition? A Case Study on CHILDES with Automated Detection of Filler-Gap Dependencies Authors: Zhenghao Herbert Zhou, William Dai, Maya Viswanathan, Simon Charlow, R. Thomas McCoy, Robert Frank Published: 2026-03-02 Categories: cs.CL arXiv: https://arxiv.org/abs/2603.02082v2 Children's acquisition of filler-gap dependencies has been argued by some to depend on innate grammatical knowledge, while others suggest that the distributional evidence available in child-directed speech suffices. Unfortunately, the relevant input is difficult to quantify at scale with fine granularity, making this question difficult to resolve. We present a system that identifies three core filler-gap constructions in spoken English corpora -- matrix wh-questions, embedded wh-questions, and relative clauses -- and further identifies the extraction site (i.e., subject vs. object vs. adjunct). Our approach combines constituency and dependency parsing, leveraging their complementary strengths for construction classification and extraction site identification. We validate the system on human-annotated data and find that it scores well across most categories. Applying the system to 57 English CHILDES corpora, we are able to characterize children's filler-gap input and their filler-gap production trajectories over the course of development, including construction-specific frequencies and extraction-site asymmetries. The resulting fine-grained labels enable future work in both acquisition and computational studies, which we demonstrate with a case study using filtered corpus training with language models. ---