Self-Improving Code Generation via Semantic Entropy and Behavioral Consensus

View PDF HTML (experimental)

Abstract:Improving the code generation capabilities of large language models (LLMs) typically relies on supervised fine-tuning or preference optimization, both of which require costly external resources such as powerful teacher models or reliable test units. However, in real-world scenarios, it is much harder to obtain reference solutions and test oracles than problem descriptions and test inputs. In this paper, we tackle a challenging yet realistic question: Can a code language model improve itself without access to a superior teacher and a test oracle? To answer this, we propose ConSelf, a self-improving approach built upon two key ideas. First, we introduce code semantic entropy, a novel metric that measures problem-level uncertainty by assessing the functional diversity of program behaviors, enabling a curriculum construction with the most learnable problems. Second, we present consensus-driven direct preference optimization (Con-DPO), a preference-based fine-tuning method that weights each preference pair by its behavioral consensus, thereby mitigating the impact of noisy self-generated supervision. Experiments on various benchmarks and backbone LLMs demonstrate that ConSelf significantly outperforms baselines, validating the effectiveness of semantic entropy-based curriculum construction and consensus-driven optimization in improving code generation without external supervision.

Comments: Accepted in the 34th IEEE/ACM International Conference on Program Comprehension (ICPC 2026)

Subjects:

Software Engineering (cs.SE); Artificial Intelligence (cs.AI); Programming Languages (cs.PL)

Cite as: arXiv:2603.29292 [cs.SE]

(or arXiv:2603.29292v1 [cs.SE] for this version)

https://doi.org/10.48550/arXiv.2603.29292

arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Wei Hu [view email] [v1] Tue, 31 Mar 2026 05:55:17 UTC (634 KB)