Paper Overview
Field: ML Authors: Zhen Zhao, Qihang Yang, Feifei Dai, Xiangfang Li, Bo Li arXiv: 2607.22471
Abstract (English)
Large Language Model (LLM)-based Test-Driven Development (TDD) has advanced automated code generation. However, existing approaches depend heavily on human-crafted test cases and cannot operate effectively when only natural-language requirements are available. Although recent work enables automatic test generation, it often overlooks the inherent stochasticity of LLMs, leading to two key defects: faulty tests generate misleading feedback that distorts code optimization, while mixed-quality test cases produce conflicting evaluation signals that hinder reliable code selection.
To address these challenges, the authors propose MineValiCoder, a collaborative closed-loop TDD framework based on the mutual reinforcement of test-case quality and code quality. MineValiCoder comprises three modules:
- Test Case Quality Mining (TCQM): filters faulty test cases via self-validation, providing reliable optimization supervision.
- Parallel TDD Refinement: iteratively optimizes code using validated test-case feedback and generates diverse, high-quality code candidates.
- Bipartite Code-Test Mutual Validation (BiCoTeV): dynamically models code-test interactions and performs mutual validation scoring for stable, reliable selection of the best code.
| Benchmark | Pass@1 | |---|---| | HumanEval | 96.34% | | MBPP | 87.40% | | APPS | 64.00% | | LiveCodeBench | 51.33% |
These results demonstrate MineValiCoder's effectiveness in mitigating LLM stochasticity and improving the reliability of automated code generation.