Hugging Face Daily Papers
500 articles archived · Visit source ↗ · RSS
-
Hugging Face Daily Papers research 9d ago
HAKARI-Bench: A Lightweight Benchmark for Comparing Retrieval Architectures and Efficiency Settings under Unified Conditions
Abstract HAKARI-Bench provides a lightweight benchmark for comparing retrieval methods across multiple configurations and languages, enabling efficient model selection and performance analysis. Generated by Qwen/Qwen2.5-Coder-32B-Instruct With the rapid spread of…
23 -
AOHP: An Open-Source OS-Level Agent Harness for Personalized, Efficient and Secure Interaction
Abstract AOHP presents an Android-based operating system framework that treats AI agents as first-class entities, enhancing task completion rates and reducing execution costs through specialized agent-oriented mechanisms. Generated by Qwen/Qwen2.5-Coder-32B-Instruct AI agents…
16 -
DataClaw0: Agentic Tailoring Multimodal Data from Raw Streams
Abstract Agentic Data Tailoring paradigm uses learnable data processing to structure high-entropy multimodal streams, with DataClaw_0-9B model achieving robust alignment through SFT and GRPO on a novel benchmark. Generated by Qwen/Qwen2.5-Coder-32B-Instruct Massive unstructured…
19 -
UniverSat: Resolution- and Modality-Agnostic Transformers for Earth Observation
Abstract UniverSat introduces a Universal Patch Encoder for Vision Transformers that enables robust, sensor-agnostic spatial feature extraction across diverse Earth Observation data types. Generated by Qwen/Qwen2.5-Coder-32B-Instruct Vision Transformers (ViT) dominate computer…
6 -
FastMix: Fast Data Mixture Optimization via Gradient Descent
Abstract FASTMIX automates optimal data mixture discovery during training by formulating mixture selection as a bilevel optimization problem that jointly optimizes mixture coefficients and model parameters through iterative updates. Generated by Qwen/Qwen2.5-Coder-32B-Instruct…
19 -
CLI-Universe: Towards Verifiable Task Synthesis Engine for Terminal Agents
Abstract A principled synthesis engine generates high-quality terminal-agent tasks through multi-dimensional capability taxonomy and evidence-guided research, creating a distilled dataset that enables significant performance gains in LLM training. Generated by…
5 -
PoLAR: Factorizing Extent and Mode in Latent Actions for Robot Policy Learning
Abstract PoLAR introduces a geometrically structured latent action representation in hyperbolic space that separates transition extent from transition mode, improving robotic policy learning performance. Generated by Qwen/Qwen2.5-Coder-32B-Instruct Latent action pretraining…
12 -
Dense Reward for Multi-View 3D Reasoning with Global Maps and Local Views
Abstract DR-MV3D presents a map-grounded learning framework with dense rewards to improve multi-view 3D visual question answering through global map construction, view-trajectory planning, and egocentric grounding. Generated by Qwen/Qwen2.5-Coder-32B-Instruct Multi-view 3D…
15 -
Causal Discovery in the Era of Agents
Abstract Language models should assist causal discovery workflows by providing contextual support and explanations rather than generating causal conclusions, as demonstrated through a platform that integrates data analysis and expert knowledge. Generated by…
31 -
EnterpriseClawBench: Benchmarking Agents from Real Workplace Sessions
Abstract EnterpriseClawBench presents a benchmark for enterprise agents based on real-world sessions with 852 reproducible tasks, emphasizing comprehensive evaluation metrics beyond single performance scores. Generated by Qwen/Qwen2.5-Coder-32B-Instruct Enterprise agents…
30 -
PolicyTrim: Boosting Intrinsic Policy Efficiency of Vision-Language-Action Models
Abstract PolicyTrim is a reinforcement learning-based framework that enhances VLA model efficiency by extending reliable action chunk lengths and reducing redundant physical steps through dynamic exploration and redundancy-aware rewards. Generated by…
25 -
Safe Few-Step Generation via Velocity Editing
Abstract VESFlow is a training-free safety method for flow matching-based text-to-image generation that edits velocity fields to ensure safe output while maintaining prompt integrity. Generated by Qwen/Qwen2.5-Coder-32B-Instruct Flow matching has recently emerged as a strong…
16 -
Tmax: A simple recipe for terminal agents
Abstract A novel RL training approach for terminal agents achieves superior performance using a simplified recipe and expanded dataset, enabling effective training with fewer parameters than previous methods. Generated by Qwen/Qwen2.5-Coder-32B-Instruct Terminal-using agents…
36 -
OpenRath: Session-Centered Runtime State for Agent Systems
Abstract OpenRath introduces a PyTorch-like programming model for multi-agent systems using Session as a central runtime abstraction that enables explicit fork, merge, and replay operations while recording comprehensive execution state. Generated by…
21 -
Learning from Your Own Mistakes: Constructing Learnable Micro-Reflective Trajectories for Self-Distillation
Abstract Trajectory-Augmented Policy Optimization (TAPO) enhances large language model reasoning by creating explicit corrective trajectories that preserve erroneous reasoning while incorporating natural-language diagnoses and corrections, outperforming traditional…
31 -
Connect the Dots: Training LLMs for Long-Lifecycle Agents with Cross-Domain Generalization Via Reinforcement Learning
Abstract Large language models can be trained through reinforcement learning to develop a meta-capability enabling continuous learning and adaptation across long sequences of tasks in dynamic environments. Generated by Qwen/Qwen2.5-Coder-32B-Instruct This work presents a general…
31 -
PlanBench-XL: Evaluating Long-Horizon Planning of LLM Tool-Use Agents in Large-Scale Tool Ecosystems
Abstract PlanBench-XL evaluates large language model agents' ability to plan and adapt in complex tool-rich environments with limited visibility and dynamic disruptions. Generated by Qwen/Qwen2.5-Coder-32B-Instruct LLM agents increasingly operate in large tool ecosystems, where…
10 -
World Action Models: A Survey
Abstract World Action Models are predictive-action systems that generate future states for decision-making, with designs balancing representational richness against computational constraints. Generated by Qwen/Qwen2.5-Coder-32B-Instruct World Action Models (WAMs) are embodied…
30 -
HydraHead: From Head-Level Functional Heterogeneity to Specialized Attention Hybridization
Abstract HydraHead is a novel attention hybridization architecture that combines Full Attention and Linear Attention at the head level, achieving superior long-context performance with reduced training overhead through interpretability-driven selection and scale-normalized…
34 -
KaLM-Reranker-V1: Fast but Not Late Interaction for Compressed Document Reranking
Abstract KaLM-Reranker-V1 is a fast reranker that decouples query and passage computation using encoder-decoder architecture with Matryoshka embedding pooling and cross-attention for efficient relevance modeling. Generated by Qwen/Qwen2.5-Coder-32B-Instruct As retrieval systems…
32 -
Manifold Bandits: Bayesian Curriculum Learning over the Latent Geometry of Large Language Models
Abstract Reinforcement learning approaches for improving LLM reasoning capabilities are enhanced by a Bayesian Manifold Curriculum framework that structures problem sampling based on task manifold relationships and endogenous non-stationarity. Generated by…
20 -
EvoEmbedding: Evolvable Representations for Long-Context Retrieval and Agentic Memory
Abstract EvoEmbedding is a dynamic embedding model that generates adaptive representations by maintaining a continuously updated latent memory, enabling improved retrieval performance in long-context scenarios. Generated by Qwen/Qwen2.5-Coder-32B-Instruct Existing embedding…
32 -
Exploring the Design Space of Reward Backpropagation for Flow Matching
Abstract FlowBP addresses limitations in flow matching model alignment by using a surrogate trajectory framework that reduces memory usage and gradient chaining while maintaining performance across multiple text-to-image models. Generated by Qwen/Qwen2.5-Coder-32B-Instruct…
23 -
DailyReport: An Open-ended Benchmark for Evaluating Search Agents on Daily Search Tasks
Abstract Search agents face challenges in real-world evaluation due to limited benchmarks and coarse metrics, necessitating more nuanced assessment approaches. Generated by Qwen/Qwen2.5-Coder-32B-Instruct Search Agents (SAs) typically leverage large language models (LLMs) to…
14 -
CalVerT: Augmenting Agents with Calibrated Verifier Telemetry Improves Action and Learning in Knowledge-Intensive Tasks
Abstract Calibrated verifier telemetry enhances LLM agents in knowledge-intensive question answering by providing confidence scores and grounding verification, reducing both over-retrieval and unsupported answers. Generated by Qwen/Qwen2.5-Coder-32B-Instruct LLM agents in…
7 -
Deep Research in Physical Sciences: A Multi-Agent Framework and Comprehensive Benchmark
Abstract PhySciBench benchmark reveals limited performance of current LLM agents in physical science research, leading to development of DelveAgent framework that improves accuracy through modular design and physics-grounded mechanisms. Generated by…
5 -
When, Where, and How: Adaptive Binning for Tabular Self-Supervised Learning
Abstract Adaptive Binning introduces a training-adaptive discretization method for self-supervised learning on medical tabular data, improving representation learning through feature-wise refinement and heterogeneous feature handling. Generated by Qwen/Qwen2.5-Coder-32B-Instruct…
13 -
Characterizing Narrative Content in Web-scale LLM Pretraining Data
Abstract A comprehensive analysis of narrative structures in large-scale language model training data reveals measurable, multidimensional narrative patterns that vary across different content sources and topics. Generated by Qwen/Qwen2.5-Coder-32B-Instruct The narrative…
21 -
SproutRAG: Attention-Guided Tree Search with Progressive Embeddings for Long-Document RAG
Abstract SproutRAG is an attention-guided hierarchical retrieval-augmented generation framework that organizes sentence-level chunks into semantically coherent units using learned inter-sentence attention, enabling multi-granularity retrieval without additional LLM calls or…
33 -
MCompassRAG: Topic Metadata as a Semantic Compass for Paragraph-Level Retrieval
Abstract MCompassRAG enhances retrieval-augmented generation by using topic-level metadata to guide chunk selection, improving both efficiency and precision in complex research tasks. Generated by Qwen/Qwen2.5-Coder-32B-Instruct Retrieval-augmented generation (RAG) systems…
32 -
StylisticBias: A Few Human Visual Cues Drive Most Social Biases in MLLMs
Abstract Multimodal large language models exhibit social bias driven by specific visual attributes, with fashion style and socioeconomic cues having the greatest impact on model judgments. Generated by Qwen/Qwen2.5-Coder-32B-Instruct Multimodal large language models (MLLMs) are…
37 -
Multi-Turn Reflective Masking Elicits Reasoning in Mask Diffusion Models
Abstract Reflective Masking enables iterative local refinement in Mask Diffusion Models through lightweight post-training, supporting multi-turn reasoning without architectural changes. Generated by Qwen/Qwen2.5-Coder-32B-Instruct While reasoning on autoregressive (AR) models is…
26 -
GeneralVLA-2: Geometry-Aware Reconstruction and Governed Memory for Robot Planning
Abstract GeneralVLA-2 addresses limitations in vision-language-action systems by introducing GeoFuse-MV3D for improved 3D reconstruction and an enhanced KnowledgeBank for better memory management in robotic manipulation tasks. Generated by Qwen/Qwen2.5-Coder-32B-Instruct…
32 -
SpatialAvatar-0: High-Quality 4D Head Avatar with Multi-Stage Reconstruction
Abstract SpatialAvatar-0 enables high-quality 4D head avatar generation by combining feed-forward prediction with per-subject refinement through a shared Gaussian representation, achieving superior performance across multiple benchmarks. Generated by…
20 -
Distilling Examples into Task Instructions: Enhanced In-Context Learning for Real-World B2B Conversations
Abstract A novel approach for B2B conversation classification that reduces token usage by 99% while improving performance and maintaining robustness as context length increases. Generated by Qwen/Qwen2.5-Coder-32B-Instruct In-context learning (ICL) is the standard method for…
8 -
GateMem: Benchmarking Memory Governance in Multi-Principal Shared-Memory Agents
Abstract Current memory agents lack reliable shared institutional deployment due to challenges in balancing utility, access control, and forgetting across multiple principals with diverse authorization contexts. Generated by Qwen/Qwen2.5-Coder-32B-Instruct Memory benchmarks for…
5 -
BrainG3N: A Dual-Purpose Tokenizer for Controllable 3D Brain MRI Generation
Abstract A 3D brain MRI generative model uses a masked-autoencoder tokenizer to create clinically informative embeddings that support both medical task performance and controlled image generation. Generated by Qwen/Qwen2.5-Coder-32B-Instruct Three-dimensional (3D) brain MRI is…
6 -
WorldLines: Benchmarking and Modeling Long-Horizon Stateful Embodied Agents
Abstract WorldLines benchmark evaluates long-term memory in embodied agents through household scenarios, while ObsMem framework addresses challenges in partial observability and memory translation for decision-making. Generated by Qwen/Qwen2.5-Coder-32B-Instruct To assist humans…
19 -
LedgerAgent: Structured State for Policy-Adherent Tool-Calling Agents
Abstract LEDGERAGENT is a method for customer service agents that maintains task states in a separate ledger to improve policy adherence and state management during tool calling. Generated by Qwen/Qwen2.5-Coder-32B-Instruct Policy-adherent tool-calling agents in customer-service…
36 -
PerceptionDLM: Parallel Region Perception with Multimodal Diffusion Language Models
Abstract PerceptionDLM enables efficient parallel region perception in multimodal diffusion language models through structured attention masking and efficient prompting, achieving faster inference without sacrificing caption quality. Generated by Qwen/Qwen2.5-Coder-32B-Instruct…
12 -
Context-Aware RL for Agentic and Multimodal LLMs
Abstract ContextRL enhances long-horizon reasoning and multimodal performance through reinforcement learning that rewards context selection for supporting query-answer pairs, achieving improvements over standard methods on diverse benchmarks. Generated by…
21 -
Freeing the Law with LOCUS: A Local Ordinance Corpus for the United States
Abstract A comprehensive corpus and access layer for U.S. local ordinance codes has been developed to enable machine-readable legal AI research, addressing the lack of authoritative legal text at scale for local regulations. Generated by Qwen/Qwen2.5-Coder-32B-Instruct Progress…
4 -
ReSyn: A Generalized Recursive Regular Expression Synthesis Framework
Abstract A divide-and-conquer framework named ReSyn enhances regex synthesis accuracy by decomposing complex problems, combined with a parameter-efficient synthesizer called Set2Regex that handles example permutation invariance. Generated by Qwen/Qwen2.5-Coder-32B-Instruct…
27 -
LegalHalluLens: Typed Hallucination Auditing and Calibrated Multi-Agent Debate for Trustworthy Legal AI
Abstract LegalHalluLens audits AI systems in legal workflows by identifying specific error patterns and directional biases in hallucinations across different claim types, enabling more reliable deployment through targeted diagnostic and mitigation approaches. Generated by…
31 -
Configurable Clinical Information Extraction with Agentic RAG: What Works, What Breaks, and Why
Abstract ACIE, an agentic RAG system deployed in a clinical setting, demonstrates high accuracy in extracting medical information from complex patient contexts, achieving 96.5% acceptance rate by nuclear-medicine physicians across 7,326 judgments. Generated by…
5 -
The Data Manifold under the Microscope
Abstract A benchmarking framework is introduced to study data-manifold geometry by extending dSprites and COIL-20 datasets with additional transformation dimensions and dense sampling, enabling accurate estimation of curvature, reach, and volume for theoretical analysis and…
36 -
The FID Lottery: Quantifying Hidden Randomness in Generative-Model Evaluation
Abstract Analysis of FID variance across different training and sampling seeds reveals significant reproducibility issues in image generation evaluation, with retraining causing larger fluctuations than resampling, and recommends updated evaluation protocols with error bars and…
21 -
Duration Aware Scheduling for ASR Serving Under Workload Drift
Abstract Duration-aware scheduling policies improve ASR serving latency by leveraging audio length as a predictor for processing time, with SJF and HRRN algorithms showing significant median latency reductions while maintaining throughput. Generated by…
26 -
Rethinking Shrinkage Bias in LLM FP4 Pretraining: Geometric Origin, Systemic Impact, and UFP4 Recipe
Abstract Uniform 4-bit training with RHT-based quantization outperforms E2M1-based methods by eliminating shrinkage bias and improving training stability across large language model architectures. Generated by Qwen/Qwen2.5-Coder-32B-Instruct FP4 training promises substantial…
31 -
Multi-LCB: Extending LiveCodeBench to Multiple Programming Languages
Abstract Multi-LCB addresses the limitation of LiveCodeBench by providing a multi-language benchmark for evaluating LLMs across twelve programming languages while maintaining contamination controls and evaluation protocols. Generated by Qwen/Qwen2.5-Coder-32B-Instruct…
33