10 papers
CamoDocs: A Poisoning Attack Against Retrieval-Augmented Language Models Using Camouflaged Documents
Jaewon Jung, Haizhong Zheng, Hongsun Jang +3
Retrieval-augmented generation (RAG) augments LLMs with external documents, but public or user-editable sources expose RAG systems to data poisoning: attackers can inject malicious…
Beyond Confidence: Test-Time Scaling for Multi-Turn Search Agents via Retrieval Grounding
Hyunho Kook, Junhyuk So, Tianyu Fu +2
Confidence-based voting aggregates parallel LLM rollouts by weighting each with internal signals such as token log probabilities, and has been actively studied for single-turn reas…
Sparrow: Sparse Rollout for Stable and Efficient Long-context RL of Large Language Models
Yang Zhou, Ranajoy Sadhukhan, Zhaofeng Sun +7
Despite being powerful, reinforcement learning with verifiable rewards (RLVR) induces extremely long COT, making it computationally expensive. Since RLVR per-step cost is dominated…
The Last Human-Written Paper: Agent-Native Research Artifacts
Jiachen Liu, Jiaxin Pei, Jintao Huang +34
Scientific publication compresses a branching, iterative research process into a linear narrative, discarding the majority of what was discovered along the way. This compilation im…
AstraFlow: Dataflow-Oriented Reinforcement Learning for Agentic LLMs
Haizhong Zheng, Yizhuo Di, Jiahui Wang +7
Reinforcement learning (RL) is increasingly used to improve the reasoning, coding, and tool-use capabilities of large language models, but agentic RL remains prohibitively expensiv…
Jackpot: Optimal Budgeted Rejection Sampling for Extreme Actor-Policy Mismatch Reinforcement Learning
Zhuoming Chen, Hongyi Liu, Yang Zhou +2
Reinforcement learning (RL) for large language models (LLMs) remains expensive, particularly because the rollout is expensive. Decoupling rollout generation from policy optimizatio…
Prosperity before Collapse: How Far Can Off-Policy RL Reach with Stale Data on LLMs?
Haizhong Zheng, Jiawei Zhao, Beidi Chen
Reinforcement learning has been central to recent advances in large language model reasoning, but most algorithms rely on on-policy training that demands fresh rollouts at every up…
When "Correct" Is Not Safe: Can We Trust Functionally Correct Patches Generated by Code Agents?
Yibo Peng, James Song, Lei Li +6
Code agents are increasingly trusted to autonomously fix bugs on platforms such as GitHub, yet their security evaluation focuses almost exclusively on functional correctness. In th…
RLBoost: Harvesting Preemptible Resources for Cost-Efficient Reinforcement Learning on LLMs
Yongji Wu, Xueshen Liu, Haizhong Zheng +5
Reinforcement learning (RL) has become essential for unlocking advanced reasoning capabilities in large language models (LLMs). RL workflows involve interleaving rollout and traini…
Kinetics: Rethinking Test-Time Scaling Laws
Ranajoy Sadhukhan, Zhuoming Chen, Haizhong Zheng +3
We rethink test-time scaling laws from a practical efficiency perspective, revealing that the effectiveness of smaller models is significantly overestimated. Prior work, grounded i…