Tag
This paper proposes ReCo, a reward-coordinated compression framework that uses a process-reward estimator to adaptively compress KV-cache, control reflection tokens, and enable early stopping, cutting generated tokens by 37–65% and latency by ~2x across reasoning models while preserving accuracy.
Proposes PaTR, a process-reward-guided adaptive tree rollout framework for multi-turn reinforcement learning in LLM agents. It selectively branches from promising states and prunes dead-end paths, achieving up to +5.0 on SWE-Bench and +9.3 on FrozenLake under the same training budget.
Introduces TRIAGE, a role-typed credit assignment framework that improves agentic reinforcement learning by providing more nuanced credit assignment than standard GRPO methods, using a structured judge to classify action segments and assign process rewards based on semantic roles.
This paper presents Process-Verified Reinforcement Learning, using the Lean proof assistant as a process oracle to provide fine-grained tactic-level feedback during training, improving theorem proving performance.
ProcessThinker introduces a practical post-training pipeline that provides step-level process rewards without training an explicit process reward model. It uses rollout-based rewards to give dense credit assignment for multi-step reasoning in multimodal LLMs, consistently improving performance on video benchmarks.
ARBOR introduces a reusable rubric buffer to provide online process rewards for LLM-based search agents, improving training efficiency when outcome-only rewards are insufficient. It outperforms GRPO and DAPO on multi-hop QA benchmarks, converting up to 42% of zero-gradient training groups into informative ones.
RoRo introduces a rubric-guided process reward framework for stepwise model routing in Large Reasoning Models, using process rewards alongside outcome rewards to train a routing policy via GRPO, outperforming baselines on reasoning benchmarks.
Introduces Metacognition-as-Reward (MaR), a reinforcement learning framework that guides LLM reasoning via metacognitive knowledge and regulation signals, achieving up to 11% improvement over vanilla methods on reasoning benchmarks.
This paper identifies and addresses the problem of 'Miracle Steps' in LLM mathematical reasoning—unjustified jumps to correct answers that indicate reward hacking—by proposing Rubric Reward Model (RRM), a process-oriented reward function that evaluates entire reasoning trajectories. RRM achieves significant improvements on AIME2024 (26.7% to 62.6% Verified Pass@1024) and reduces Miracle Steps by 71%.