Revisiting LLM Reasoning via Information Bottleneck
Fuente:
arXiv
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866908464611065856 |
|---|---|
| author | Lei, Shiye Cheng, Zhihao Jia, Kai Tao, Dacheng |
| author_facet | Lei, Shiye Cheng, Zhihao Jia, Kai Tao, Dacheng |
| contents | Large language models (LLMs) have recently demonstrated remarkable progress in reasoning capabilities through reinforcement learning with verifiable rewards (RLVR). By leveraging simple rule-based rewards, RL effectively incentivizes LLMs to produce extended chain-of-thought (CoT) reasoning trajectories, progressively guiding them toward correct answers. However, existing approaches remain largely heuristic and intuition-driven, limiting the development of principled methodologies. In this paper, we present a theoretical characterization of LLM reasoning grounded in information bottleneck (IB) principle, introducing IB-aware reasoning optimization (IBRO), a framework that encourages reasoning trajectories to be both informative about the final correct answer and generalizable across diverse prompts. We derive a practical token-level surrogate objective and propose an efficient approximation, resulting in the lightweight IB regularization method. This technique integrates seamlessly into existing RL-based post-training frameworks without additional computational overhead, requiring only a one-line code modification. Empirically, we validate IB regularization across multiple mathematical reasoning benchmarks and RL algorithms, demonstrating consistent improvements in LLM reasoning performance. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2507_18391 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Revisiting LLM Reasoning via Information Bottleneck Lei, Shiye Cheng, Zhihao Jia, Kai Tao, Dacheng Artificial Intelligence Large language models (LLMs) have recently demonstrated remarkable progress in reasoning capabilities through reinforcement learning with verifiable rewards (RLVR). By leveraging simple rule-based rewards, RL effectively incentivizes LLMs to produce extended chain-of-thought (CoT) reasoning trajectories, progressively guiding them toward correct answers. However, existing approaches remain largely heuristic and intuition-driven, limiting the development of principled methodologies. In this paper, we present a theoretical characterization of LLM reasoning grounded in information bottleneck (IB) principle, introducing IB-aware reasoning optimization (IBRO), a framework that encourages reasoning trajectories to be both informative about the final correct answer and generalizable across diverse prompts. We derive a practical token-level surrogate objective and propose an efficient approximation, resulting in the lightweight IB regularization method. This technique integrates seamlessly into existing RL-based post-training frameworks without additional computational overhead, requiring only a one-line code modification. Empirically, we validate IB regularization across multiple mathematical reasoning benchmarks and RL algorithms, demonstrating consistent improvements in LLM reasoning performance. |
| title | Revisiting LLM Reasoning via Information Bottleneck |
| topic | Artificial Intelligence |
| url | https://arxiv.org/abs/2507.18391 |