papers.joyehuang.me

Paper Recommendations

Daily AI-powered curation from arXiv and Hugging Face Daily Papers. Last updated May 09, 2026 · 23:01 UTC.

Archive joyehuang.me Source
77
Candidates
5
Top Picks
0
Blockbusters
13
Max Score
Today's Picks

Editor's Rationale

MASPO: Joint Prompt Optimization for LLM-based Multi-Agent Systems Multi-Agent prompt joint optimization framework. Core innovation is the joint evaluation mechanism - not only evaluating a single agent's prompt quality, but also assessing its facilitation effect on downstream successor agents, bridging the gap between local interaction and global goals. Combined with data-driven evolutionary beam search, achieves average 2.9% improvement across 6 tasks. Very directly useful for anyone doing multi-agent orchestration. STALE: Can LLM Agents Know When Their Memories Are No Longer Valid? A critical blind spot in agent long-term memory management: Implicit Conflict - new observations indirectly invalidate old memories without explicit negation. STALE constructs 400 expert-validated conflict scenarios (up to 150K tokens context), probing from three dimensions: State Resolution, Premise Resistance, and Implicit Policy Adaptation. Best model only achieves 55.2%, showing current agents are still weak at memory staleness awareness. Worth reading for anyone designing context/memory systems; includes CUPMem baseline. SkillOS: Learning Skill Curation for Self-Evolving Agents Agent self-evolution skill management problem. Existing methods either require manual maintenance, heuristic rules, or only optimize short-horizon operations. SkillOS uses RL to train a skill curator (separate from frozen executor), learning long-term skill CRUD from indirect delayed feedback. Valuable reference for designing agent systems that can continuously learn and accumulate reusable experience. AI Co-Mathematician: Accelerating Mathematicians with Agentic AI DeepMind AI math assistant workbench. Though domain-specific to mathematics, the system architecture is highly inspiring: asynchronous, stateful workspace managing uncertainty, refining user intent, tracking failed hypotheses, and outputting native mathematical artifacts. This human-collaborative-workflow design paradigm is instructive for any agent product requiring long-cycle, open-ended exploration. Achieves 48% on FrontierMath Tier 4 (highest among all AI systems).

Top Picks

5 papers
#01 13

AI Co-Mathematician: Accelerating Mathematicians with Agentic AI

Daniel Zheng, Ingrid von Glehn, Yori Zwols et al.
cs.AI
We introduce the AI co-mathematician, a workbench for mathematicians to interactively leverage AI agents to pursue open-ended research. The AI co-mathematician is optimized to provide holistic support for the exploratory and iterative reality of mathematical workflows, including …
agentagenticworkflowRAGbenchmarkPPO
#02 12

Patch2Vuln: Agentic Reconstruction of Vulnerabilities from Linux Distribution Binary Patches

Isaac David, Arthur Gervais
cs.CR, cs.AI
Security updates create a short but important window in which defenders and attackers can compare vulnerable and patched software. Yet in many operational settings, the most accessible artifacts are binary packages rather than source patches or advisory text. This paper asks whet…
agentagenticreasoningRAGPPO
#03 11

MASPO: Joint Prompt Optimization for LLM-based Multi-Agent Systems

Zhexuan Wang, Xuebo Liu, Li Wang et al.
cs.AI, cs.CL
Large language model (LLM)-based Multi-agent systems (MAS) have shown promise in tackling complex collaborative tasks, where agents are typically orchestrated via role-specific prompts. While the quality of these prompts is pivotal, jointly optimizing them across interacting agen…
agentmulti-agentRAGevaluationalignment
#04 11

STALE: Can LLM Agents Know When Their Memories Are No Longer Valid?

Hanxiang Chao, Yihan Bai, Rui Sheng et al.
cs.CL
Large Language Model (LLM) agents are increasingly expected to maintain coherent, long-term personalized memory, yet current benchmarks primarily measure static fact retrieval, overlooking the ability to revise stored beliefs when new evidence emerges. We identify a critical and …
agentagenticreasoningevaluationbenchmarkPPO
#05 11

ResRL: Boosting LLM Reasoning via Negative Sample Projection Residual Reinforcement Learning

Zihan Lin, Xiaohan Wang, Jie Cao et al.
Reinforcement Learning with Verifiable Rewards (RLVR) enhances reasoning of Large Language Models (LLMs) but usually exhibits limited generation diversity due to the over-incentivization of positive rewards. Although methods like Negative Sample Reinforcement (NSR) mitigate this …
agentfunction callingreasoningRAGbenchmarkalignment

All Candidates

77 papers
ScorePaperAuthorsCategory
10 Audio-Visual Intelligence in Large Foundation Models You Qin, Kai Liu et al.
10 RemoteZero: Geospatial Reasoning with Zero Human Annotations Liang Yao, Fan Liu et al.
9 StraTA: Incentivizing Agentic Reinforcement Learning with Strategic Trajectory Abstraction Xiangyuan Xue, Yifan Zhou et al. cs.CL, cs.AI
9 Cited but Not Verified: Parsing and Evaluating Source Attribution in LLM Deep Research Agents Hailey Onweller, Elias Lumer et al. cs.CL
9 Turning Drift into Constraint: Robust Reasoning Alignment in Non-Stationary Environments Xiaoyu Yang, En Yu et al.
8 Superintelligent Retrieval Agent: The Next Frontier of Information Retrieval Zeyu Yang, Qi Ma et al. cs.IR, cs.AI, cs.LG
8 Can RL Teach Long-Horizon Reasoning to LLMs? Expressiveness Is Key Tianle Wang, Zhaoyang Wang et al. cs.AI, cs.CL
8 SkillOS: Learning Skill Curation for Self-Evolving Agents Siru Ouyang, Jun Yan et al. cs.AI, cs.CL
8 AI CFD Scientist: Toward Open-Ended Computational Fluid Dynamics Discovery with Physics-Aware AI Agents Nithin Somasekharan, Rabi Pathak et al. physics.flu-dyn, cs.AI
8 SWE-WebDevBench: Evaluating Coding Agent Application Platforms as Virtual Software Agencies Siddhant Saxena, Nilesh Trivedi et al.
7 Verifier-Backed Hard Problem Generation for Mathematical Reasoning Yuhang Lai, Jiazhan Feng et al. cs.LG, cs.AI, cs.CL
7 BioTool: A Comprehensive Tool-Calling Dataset for Enhancing Biomedical Capabilities of Large Language Models Xin Gao, Ruiyi Zhang et al.
6 The Scaling Properties of Implicit Deductive Reasoning in Transformers Enrico Vompa, Tanel Tammet
6 Balanced Aggregation: Understanding and Fixing Aggregation Bias in GRPO Zhiyuan Zeng, Jiameng Huang et al.
6 Auto Research with Specialist Agents Develops Effective and Non-Trivial Training Recipes Jingjie Ning, Xiaochuan Li et al.
5 Are We Making Progress in Multimodal Domain Generalization? A Comprehensive Benchmark Study Hao Dong, Hongzhao Li et al. cs.CV, cs.AI, cs.LG, cs.MM
5 GlazyBench: A Benchmark for Ceramic Glaze Property Prediction and Image Generation Ziyu Zhai, Siyou Li et al. cs.AI, cs.CV
5 Recursive Agent Optimization Apurva Gandhi, Satyaki Chakraborty et al. cs.LG, cs.AI, cs.CL, cs.MA
5 Beyond Negative Rollouts: Positive-Only Policy Optimization with Implicit Negative Gradients Mingwei Xu, Hao Fang cs.CL
5 KernelBench-X: A Comprehensive Benchmark for Evaluating LLM-Generated GPU Kernels Han Wang, Jintao Zhang et al.
5 EDU-CIRCUIT-HW: Evaluating Multimodal Large Language Models on Real-World University-Level STEM Student Handwritten Solutions Weiyu Sun, Liangliang Chen et al.
5 TabEmbed: Benchmarking and Learning Generalist Embeddings for Tabular Understanding Minjie Qiang, Mingming Zhang et al.
5 Nonsense Helps: Prompt Space Perturbation Broadens Reasoning Exploration Langlin Huang, Chengsong Huang et al.
5 MiA-Signature: Approximating Global Activation for Long-Context Understanding Yuqing Li, Jiangnan Li et al.
5 XL-SafetyBench: A Country-Grounded Cross-Cultural Benchmark for LLM Safety and Cultural Sensitivity Dasol Choi, Eugenia Kim et al.
4 BAMI: Training-Free Bias Mitigation in GUI Grounding Borui Zhang, Bo Zhang et al. cs.CV, cs.AI
4 Parser agreement and disagreement in L2 Korean UD: Implications for human-in-the-loop annotation Hakyung Sung, Gyu-Ho Shin cs.CL
4 PairAlign: A Framework for Sequence Tokenization via Self-Alignment with Applications to Audio Tokenization Adhiraj Banerjee, Vipul Arora cs.LG, cs.CL, cs.SD
4 Generative Quantum-inspired Kolmogorov-Arnold Eigensolver Yu-Cheng Lin, Yu-Chao Hsu et al.
4 When to Trust Imagination: Adaptive Action Execution for World Action Models Rui Wang, Yue Zhang et al.
4 Think, then Score: Decoupled Reasoning and Scoring for Video Reward Modeling Yuan Wang, Ouxiang Li et al.
4 Skill1: Unified Evolution of Skill-Augmented Agents via Reinforcement Learning Yaorui Shi, Yuxin Chen et al.
4 MiniCPM-o 4.5: Towards Real-Time Full-Duplex Omni-Modal Interaction Junbo Cui, Bokai Xu et al.
3 When No Benchmark Exists: Validating Comparative LLM Safety Scoring Without Ground-Truth Labels Sushant Gautam, Finn Schwall et al. cs.LG, cs.AI, cs.CL
3 Edge-specific signal propagation on mature chromophore-region 3D mechanism graphs for fluorescent protein quantum-yield prediction Yuchen Xiong, Swee Keong Yeap et al. cs.LG
3 UniSD: Towards a Unified Self-Distillation Framework for Large Language Models Yiqiao Jin, Yiyang Wang et al. cs.CL, cs.AI, cs.LG
3 Long Context Pre-Training with Lighthouse Attention Bowen Peng, Subho Ghosh et al. cs.CL
3 RaguTeam at SemEval-2026 Task 8: Meno and Friends in a Judge-Orchestrated LLM Ensemble for Faithful Multi-Turn Response Generation Ivan Bondarenko, Roman Derunets et al.
3 ReflectDrive-2: Reinforcement-Learning-Aligned Self-Editing for Discrete Diffusion Driving Huimin Wang, Yue Wang et al.
3 The Granularity Axis: A Micro-to-Macro Latent Direction for Social Roles in Language Models Chonghan Qin, Xiachong Feng et al.
3 A Foundation Model for Zero-Shot Logical Rule Induction Yin Jun Phua
2 ActCam: Zero-Shot Joint Camera and 3D Motion Control for Video Generation Omar El Khalifi, Thomas Rossi et al. cs.CV, cs.AI, cs.LG
2 Optimizer-Model Consistency: Full Finetuning with the Same Optimizer as Pretraining Forgets Less Yuxing Liu, Jianyu Wang et al. cs.LG, cs.AI, math.OC
2 PianoCoRe: Combined and Refined Piano MIDI Dataset Ilya Borovik cs.SD, cs.LG
2 EMO: Pretraining Mixture of Experts for Emergent Modularity Ryan Wang, Akshita Bhagia et al. cs.CL
2 Continuous Latent Diffusion Language Model Hongcan Guo, Qinyu Zhao et al. cs.CL, cs.AI, cs.CV
2 Sparkle: Realizing Lively Instruction-Guided Video Background Replacement via Decoupled Guidance Ziyun Zeng, Yiqi Lin et al.
2 MARBLE: Multi-Aspect Reward Balance for Diffusion RL Canyu Zhao, Hao Chen et al.
2 The First Token Knows: Single-Decode Confidence for Hallucination Detection Mina Gabriel
1 When and Why SignSGD Outperforms SGD: A Theoretical Study Based on $\ell_1$-norm Lower Bounds Hongyi Tao, Dingzhi Yu et al. cs.LG, cs.AI, cs.CL, math.OC
1 Why Global LLM Leaderboards Are Misleading: Small Portfolios for Heterogeneous Supervised ML Jai Moondra, Ayela Chughtai et al. cs.LG, cs.DM, cs.ET, math.OC
1 Crafting Reversible SFT Behaviors in Large Language Models Yuping Lin, Pengfei He et al. cs.LG
1 Online Bayesian Calibration under Gradual and Abrupt System Changes Yang Xu, Chiwoo Park cs.LG, cs.ET, stat.ML
1 Algospeak, Hiding in the Open: The Trade-off Between Legible Meaning and Detection Avoidance Jan Fillies, Ronald E. Robertson et al. cs.CL, cs.CY
1 Automated Clinical Report Generation for Remote Cognitive Remediation: Comparing Knowledge-Engineered Templates and LLMs in Low-Resource Settings Yongxin Zhou, Fabien Ringeval et al. cs.CL
1 Efficient Pre-Training with Token Superposition Bowen Peng, Théo Gigant et al. cs.CL
1 TIDE: Every Layer Knows the Token Beneath the Context Ajay Jaiswal, Lauren Hannah et al.
1 Diffusion Model as a Generalist Segmentation Learner Haoxiao Wang, Antao Xiang et al.
0 UniPool: A Globally Shared Expert Pool for Mixture-of-Experts Minbin Huang, Han Shi et al. cs.LG, cs.AI
0 Concept-Based Abductive and Contrastive Explanations for Behaviors of Vision Models Ronaldo Canizales, Divya Gopinath et al. cs.LG, cs.AI
0 The Structural Origin of Attention Sink: Variance Discrepancy, Super Neurons, and Dimension Disparity Siquan Li, Kaiqi Jiang et al. cs.LG, cs.AI, stat.ML
0 Inductive Venn-Abers and related regressors Ivan Petej, Vladimir Vovk cs.LG
0 Hybrid Quantum-Classical GANs for the Generation of Adversarial Network Flows Prateek Paudel, Nitin Jha et al. cs.LG
0 LiVeAction: a Lightweight, Versatile, and Asymmetric Neural Codec Design for Real-time Operation Dan Jacobellis, Neeraja J. Yadwadkar eess.IV, cs.LG, cs.MM, eess.AS, eess.SP
0 SoftSAE: Dynamic Top-K Selection for Adaptive Sparse Autoencoders Jakub Stępień, Marcin Mazur et al. cs.LG, cs.CV
0 GeoStack: A Framework for Quasi-Abelian Knowledge Composition in VLMs Pranav Mantini, Shishir K. Shah
0 Prescriptive Scaling Laws for Data Constrained Training Justin Lovelace, Christian Belardi et al.
0 A^2TGPO: Agentic Turn-Group Policy Optimization with Adaptive Turn-level Clipping Dingwei Chen, Zefang Zong et al.
0 SwiftI2V: Efficient High-Resolution Image-to-Video Generation via Conditional Segment-wise Generation YaoYang Liu, Yuechen Zhang et al.
0 Recovering Hidden Reward in Diffusion-Based Policies Yanbiao Ji, Qiuchang Li et al.
0 TT4D: A Pipeline and Dataset for Table Tennis 4D Reconstruction From Monocular Videos Nima Rahmanian, Daniel Kienzle et al.
-1 Continuous-Time Distribution Matching for Few-Step Diffusion Distillation Tao Liu, Hao Yan et al.