Skip to main content
Search

# Temporal Credit Assignment

2 related articles

Subscribe to updates
  1. arXiv

    Reward no longer goes only to the nearest action: letting a language model explain outcomes traces delayed credit to its source

    A credit-assignment framework that redistributes delayed reward by explanation rather than temporal proximity traces sparse terminal reward back to the actions that actually mattered hundreds of steps earlier, across Crafter, WordCraft, KuaiRand, and an industry-scale generative recommender.

    University of Cambridge
    Oriol Corcoll（First author）
    Explanatory World ModelsTemporal Credit AssignmentConservative Hierarchical Redistribution OperatorReinforcement LearningRecommender systems
  2. arXiv

    DCSD decouples credit direction from magnitude in self-distillation, topping 11 benchmarks and lifting math reasoning by 8.45 points over base models

    The work introduces Decoupled Credit Self-Distillation (DCSD), which uses belief-margin probing to set each reasoning step's credit direction and marginal information gain to quantify its contribution magnitude, leaving the privileged teacher to allocate credit only within steps; across 11 mathematical and multimodal reasoning benchmarks DCSD achieves the best overall scores against GRPO, OPSD, RLSD and RLCSD, improving overall score by 8.45 points on mathematical reasoning and 7.01 points on multimodal reasoning over base models, while correcting credit direction for about 6% of tokens and reducing token credit magnitude by roughly 1.5 times.

    University of North TexasAdelaide UniversityCurtin UniversityThe University of Sydney
    Yugu Li（First author）
    Decoupled Credit Self-DistillationBelief-Margin ProbingMarginal Information GainOn-Policy Self-DistillationTemporal Credit Assignment

You’re all caught up