Episode

DelTA: Discriminative Token Credit Assignment for Reinforcement Learning from Verifiable Rewards

Podcast
Daily Paper Cast
Published
May 23, 2026
Duration seconds
1286
Processing state
not_requested
Canonical source
https://share.transistor.fm/s/428a7c58
Audio
https://media.transistor.fm/428a7c58/929d124c.mp3
JSON
/v1/public/podcasts/daily-paper-cast-7079649/episodes/delta-discriminative-token-credit-assignment-for-reinforcement-learning-from-verifiable-rewards
Markdown
/podcast/daily-paper-cast-7079649/delta-discriminative-token-credit-assignment-for-reinforcement-learning-from-verifiable-rewards.md

Actions

  • POST https://stenobird.com/v1/public/podcasts/daily-paper-cast-7079649/episodes/delta-discriminative-token-credit-assignment-for-reinforcement-learning-from-verifiable-rewards/transcription-requests
    Idempotently request low-priority transcript generation for this episode.
  • GET https://stenobird.com/podcast/daily-paper-cast-7079649/delta-discriminative-token-credit-assignment-for-reinforcement-learning-from-verifiable-rewards.md
    Read the agent-friendly Markdown representation of this episode resource.

Summary

🤗 Upvotes: 124 | cs.LG, cs.CL Authors: Kaiyi Zhang, Wei Wu, Yankai Lin Title: DelTA: Discriminative Token Credit Assignment for Reinforcement Learning from Verifiable Rewards Arxiv: http://arxiv.org/abs/2605.21467v1 Abstract: Reinforcement learning from verifiable rewards (RLVR) has emerged as a central technique for improving the reasoning capabilities of large language models. Despite its effectiveness, how response-level rewards translate into token-level probability changes remains poorly understood. We introduce a discriminator view of RLVR updates, showing that the policy-gradient update direction implicitly acts as a linear discriminator over token-gradient vectors and thereby determines which token probabilities are increased or decreased during learning. Under standard sequence-level RLVR, this discriminator is constructed from positive- and negative-side centroids formed by advantage-weighted averaging of token-gradient vectors. However, such centroid construction can be dominated by shared high-frequency patterns, such as formatting tokens, diluting sparse yet discriminative directions that better distinguish high-reward responses from low-reward ones. To address this limitation, we propose $\textbf{DelTA}$, a discriminative token credit assignment method that estimates token coefficients to amplify side-specific token-gradient directions and downweight shared or weakly discriminative ones. These coefficients reweight a self-normalized RLVR surrogate, making the effective side-wise centroids more contrastive and thereby reshaping the RLVR update direction. On seven mathematical benchmarks, DelTA outperforms the strongest same-scale baselines by 3.26 and 2.62 average points on Qwen3-8B-Base and Qwen3-14B-Base, respectively. Additional results on code generati…