Papers
Topics
Authors
Recent
Search
2000 character limit reached

Information Bottleneck Tree-based Policy Optimization

Updated 5 July 2026
  • The paper introduces IB-TPO, an online RL framework using a step-wise Information Bottleneck objective and IB-Score to balance exploration and exploitation in LLM reasoning.
  • It employs an IB-guided tree sampling method that reuses shared prefixes to improve Monte Carlo estimation, reduce token redundancy, and enhance sampling efficiency.
  • Empirical results show that IB-TPO outperforms standard GRPO methods by maintaining stable IB-Score, better exploration–exploitation balance, and improved reasoning quality.

66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 OR ti:\666 66 66 66 66 66 66 66 66 66 OR ti:\666 66 OR ti:\666 66 66 66 66 66 66 66 66 66 OR ti:\666 66 66 66 66 66 66 66 66 OR ti:\666 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66^ 66 66 OR ti:\666 66 66 66 66 66 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66^ 66 66 66 66 66 66 66 66 66^ 66 66 66 66 66 66 66 66 66 66 66^ 66 66^ 66 66 66 66 66 66 66 66 OR ti:\666 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 OR ti:\666 66 66 66 66^ 66 66^ 66 66 66 66^ 66 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 OR ti:\666 66 66 66 66 66 66 66 66 66 66 66 66 OR ti:\666 OR ti:\666 66^ 66 OR ti:\666 66^ 66 66 66 66 66 66 66 66^ 66 OR ti:\666 66^ 66 66^ 66 OR ti:\666 66^ 66 66 66 66 66 66 66 66 66 66^ 66 OR ti:\666 OR ti:\666 66^ 66 66 66 66 OR ti:\666 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 OR ti:\666 66 66 66 66 66 66 66 66 66 66 66 66 OR ti:\666 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 OR ti:\666 66^ 66 66 66 66 66 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66^ 66 66^ 66 66^ 66 OR ti:\666 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66^ 66 66 66 66 66 66 66 66^ 66 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66^ 66 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 OR ti:\666 66 66 66 66 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66^ 66 66 66 66 66 66 66 66 66 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 OR ti:\666 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66^ 66 OR ti:\666 66^ 66 66 66 66 66 66^ 66 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 OR ti:\666 66 66 66 66 66 66 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 OR ti:\666 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66^ 66 66 OR ti:\666 66 66 66 66 66 66^ 66 66 66 66 66 OR ti:\666 66 66 66 66 OR ti:\666 66 66 66 66 66 66 66 66 OR ti:\666 66 OR ti:\666 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66^ 66 66 66 66 66 66 66 66 66 66 66 OR ti:\666 66 66 66 66 OR ti:\666 66 66 66 OR ti:\666 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 OR ti:\666 66 66 OR ti:\666 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 OR ti:\666 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 OR ti:\666 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66^ 66 66 66 66 66 66 66 66 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 OR ti:\666 66 66 OR ti:\666 OR ti:\666 66 66 66 OR ti:\666 66 66 66 OR ti:\666 66 66 66 66 66 66 OR ti:\666 66 66 66 66 OR ti:\666 66 66 66 66 OR ti:\666 OR ti:\666 66 66^ 66 66 66 66 66 66 66 66 66 66 66 OR ti:\666 66 66 66 66 66 OR ti:\666 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 OR ti:\666 66 66 66 66 OR ti:\666 66 66 66 66 66 66 66 66 OR ti:\666 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 OR ti:\666 66 66 66 66 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66) OR ti:\666 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66

Definition Search Book Streamline Icon: https://streamlinehq.com
References (2)

Topic to Video (Beta)

No one has generated a video about this topic yet.

Sign Up to Generate All Videos Subscribe on YouTube

Whiteboard

No one has generated a whiteboard explanation for this topic yet.

Sign Up to Generate

Follow Topic

Get notified by email when new papers are published related to Information Bottleneck-driven Tree-based Policy Optimization (IB-TPO).

Sign Up to Follow Topic by Email