…A lightweight probe predicts the expected verifiable reward from the hidden states of the prompt and generated trajectory, as well as token-entropy statistics, and is trained online alongside the policy. To…
Papers arxiv:2605.10889 Unmasking On-Policy Distillation: Where It Helps, Where It Hurts, and Why Published on May 11 Submitted by Mohammadreza Armandpour on May 12 Apple Authors: , , , , , , , , Abstract A training…
…Through parameter-level and theoretical analyses, we show that RL induces smaller, sparser, and approximately orthogonal updates across tasks: SFT interference is norm-limited, whereas RL interference is variance-limited. Based on…
…when to update their state, when to preserve their state, and what to ignore. We study this challenge as Contextual Belief Management (CBM): maintaining a predicted belief state aligned with formal evidence…
…From Imitation to Reward Fine-Tuning with Unbiased Group Advantages and Dynamic Coefficient Rectification (2026) Bridging SFT and RL: Dynamic Policy Optimization for Robust Reasoning (2026) GAGPO: Generalized Advantage Grouped Policy Optimization…
…Siyin Wang , , , , , Jingjing Gong , Abstract ICWM enables robot policies to infer system variables from self-generated interactions, allowing adaptation to novel configurations without parameter updates by treating system identification as an in…
…both the original credit budget and verifier-determined direction. On long-CoT mathematical reasoning benchmarks, CSCR consistently outperforms GRPO baseline with the same number of policy updates. Targeted ablations further corroborate our…
…policies that bake competence into frozen weights (vision-language-action, or VLA, models), and agents that write and refine their own executable skills as code. This survey organises the field around that…
…On the system side, an asynchronous design overlaps rollout , gradient update , and policy refresh across iterations, paired with two web-agent-specific adaptations, namely an everlasting rollout pool and lightweight screenshot handling…
…an RL agent reads streaming rate and signal-sensitive features and updates trigger thresholds to maximize signal efficiency while holding the background rate inside a tolerance band. We adapt Group-Filtered Policy…