Tag
The paper proposes DEEPO, a dual-stage reinforcement learning optimization method to reduce hallucination in multimodal large language models by addressing weaknesses in the correction chain from reward to parameter update.
This paper introduces Reinforcement Learning with Decomposed Subtasks (RLDS), a method that decomposes trajectory reward into per-subtask advantages to improve credit assignment in reinforcement learning for language model agents, showing significant gains on high-heterogeneity agentic benchmarks.
This paper introduces SCOPE, a joint training method for computer-use agents that improves both task completion and safety, using a synthesized dataset and achieving strong performance on benchmarks OSWorld and OS-BLIND.
This paper introduces GVPO++, a novel post-training method for large language models that integrates KL-constrained reward maximization, enhancing training stability and enabling on-policy distillation.
DAPO is an open-source reinforcement learning system for large language models, developed by ByteDance Seed and Tsinghua AIR, which achieves state-of-the-art performance on the AIME 2024 benchmark with the Qwen2.5-32B model.
RewardVerse presents a rubric-based framework for video reward modeling to mitigate scalar drift and provide stable evaluation criteria, enhancing reinforcement learning in video generation.
The paper introduces BATON, a dual-axis policy optimization framework for LLM agents using Bayesian Feedback Attribution and Trajectory Mass Normalization, demonstrating improved performance in reinforcement learning experiments.
This research paper explores using a reference policy to deliberately steer equilibrium selection in regularized self-play for two-player zero-sum games, providing experimental results and theoretical analysis on tractable games.
This paper proposes PIVOT, a dual-level learning framework that enhances visually-grounded reasoning in large vision-language models by using self-calibrated experience replay and vision-guided advantage allocation to optimize reinforcement learning.
TIAO is a token importance-aware reinforcement learning strategy for text summarization that reweights trajectories based on token dependencies, achieving results comparable to GPT-4 and GPT-5-nano on real-world datasets.
EvolveTrade introduces a self-evolving framework for LLM trading agents that refines tool-use policies using decision traces and portfolio feedback, improving Sharpe Ratio and cumulative returns across market regimes.
Bellman Policy Optimization (BPO) is a critic-free reinforcement learning method that reformulates Policy Mirror Descent using the Bellman equation for autoregressive generation with terminal rewards, improving mathematical reasoning in large language models.
The paper introduces FGPO, a method for exact policy optimization in genomic tool selection that enumerates all possible tool subsets to outperform sampling-based approaches like GRPO.
VERPO introduces a framework for verified evidence regularized policy optimization in language models, improving performance on scientific reasoning and tool-use tasks by treating evidence as a proposal for policy correction while maintaining outcome objectives.
The paper proposes DATPO, a difficulty-adaptive tree-structured policy optimization method that enhances reasoning coverage in large models through sentence-entropy-guided branching and diversity-aware optimization, outperforming baselines in pass@k metrics.
A developer discusses why tail reinforcement learning remains effective even when the initial policy lacks good coverage of target behaviors.
TIGPO proposes a temporal instance-graph policy optimization method that extends graph-based credit assignment across policy updates for long-horizon LLM agents, using persistent transition graphs and revisit slots to improve advantage estimation and performance on benchmarks like ALFWorld and WebShop.
DMRL is a Document-Mediated Reinforcement Learning framework that optimizes skill documents for advertising recommendation using structured editing actions, policy optimization, and long-term reward prediction, demonstrating improved performance over baselines in large-scale deployment.
PGPO proposes potential-guided policy optimization for multi-turn agentic tasks, enabling finer-grained credit assignment in LLM post-training and showing strong results on ALFWorld and WebShop benchmarks.
The paper proposes Group Adaptive Clustering Policy Optimization (GAPO), a plug-in modification to GRPO methods that adapts the clipping boundary to rollout advantage, improving Pass@1 and Pass@k on math reasoning and coding benchmarks.