Self-Training Large Language Models with Confident Reasoning
This work addresses the challenge of reducing human supervision costs for training LLMs on reasoning tasks, though it is incremental as it builds on existing self-training approaches.
The paper tackles the problem of self-training large language models for reasoning by addressing the limitation of existing methods that focus only on answer confidence, proposing a new method that uses reasoning-level confidence to select high-quality reasoning paths for fine-tuning, resulting in improved accuracy on multiple benchmarks.
Large language models (LLMs) have shown impressive performance by generating reasoning paths before final answers, but learning such a reasoning path requires costly human supervision. To address this issue, recent studies have explored self-training methods that improve reasoning capabilities using pseudo-labels generated by the LLMs themselves. Among these, confidence-based self-training fine-tunes LLMs to prefer reasoning paths with high-confidence answers, where confidence is estimated via majority voting. However, such methods exclusively focus on the quality of the final answer and may ignore the quality of the reasoning paths, as even an incorrect reasoning path leads to a correct answer by chance. Instead, we advocate the use of reasoning-level confidence to identify high-quality reasoning paths for self-training, supported by our empirical observations. We then propose a new self-training method, CORE-PO, that fine-tunes LLMs to prefer high-COnfidence REasoning paths through Policy Optimization. Our experiments show that CORE-PO improves the accuracy of outputs on four in-distribution and two out-of-distribution benchmarks, compared to existing self-training methods.