Beyond Uniform Credit: Causal Credit Assignment for Policy Optimization

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Khandoga, Mykola, Yuan, Rui, Sankarapu, Vinay Kumar
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866918330560937984
author Khandoga, Mykola
Yuan, Rui
Sankarapu, Vinay Kumar
author_facet Khandoga, Mykola
Yuan, Rui
Sankarapu, Vinay Kumar
contents Policy gradient methods for language model reasoning, such as GRPO and DAPO, assign uniform credit to all generated tokens - the filler phrase "Let me think" receives the same gradient update as the critical calculation "23 + 45 = 68." We propose counterfactual importance weighting: mask reasoning spans, measure the drop in answer probability, and upweight tokens accordingly during policy gradient updates. Our method requires no auxiliary models or external annotation, instead importance is estimated directly from the policy model's own probability shifts. Experiments on GSM8K across three models spanning the Qwen and Llama families demonstrate consistent improvements over uniform baselines and faster convergence to equivalent accuracy. Inverting the importance signal hurts performance, confirming we capture genuine causal structure rather than noise. Analysis shows the method correctly prioritizes calculation steps over scaffolding text. We view these findings as establishing counterfactual importance weighting as a foundation for further research rather than a complete solution.
format Preprint
id arxiv_https___arxiv_org_abs_2602_09331
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Beyond Uniform Credit: Causal Credit Assignment for Policy Optimization
Khandoga, Mykola
Yuan, Rui
Sankarapu, Vinay Kumar
Computation and Language
Artificial Intelligence
Machine Learning
Policy gradient methods for language model reasoning, such as GRPO and DAPO, assign uniform credit to all generated tokens - the filler phrase "Let me think" receives the same gradient update as the critical calculation "23 + 45 = 68." We propose counterfactual importance weighting: mask reasoning spans, measure the drop in answer probability, and upweight tokens accordingly during policy gradient updates. Our method requires no auxiliary models or external annotation, instead importance is estimated directly from the policy model's own probability shifts. Experiments on GSM8K across three models spanning the Qwen and Llama families demonstrate consistent improvements over uniform baselines and faster convergence to equivalent accuracy. Inverting the importance signal hurts performance, confirming we capture genuine causal structure rather than noise. Analysis shows the method correctly prioritizes calculation steps over scaffolding text. We view these findings as establishing counterfactual importance weighting as a foundation for further research rather than a complete solution.
title Beyond Uniform Credit: Causal Credit Assignment for Policy Optimization
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2602.09331