Accepted Works
PARNESS 2026 — All accepted works sorted by session.
AutoFL: Automated Fault Localization via LLM-Driven Test Generation
Yuchen Wang, Jianfei Yu, Yuanjue Fan
Peking University / Tsinghua University
We propose AutoFL, a fully automated fault localization technique that leverages large language models to generate targeted test cases for pinpointing software bugs. Our approach achieves state-of-the-art results on the Defects4J benchmark without any manual intervention.
NeuralArchitecture-Bench: A Fully Automated Benchmark for NAS Evaluation
Yujie Fei, Zheyuan Zhang, Yuchen Wang
Peking University / MIT
We present NeuralArchitecture-Bench, a fully automated benchmarking framework for Neural Architecture Search (NAS). Given only a dataset URL, our system automatically designs, trains, and evaluates candidate architectures, producing comprehensive comparison reports.
AutoSurvey: LLM-Driven Automated Literature Review Generation
Zheyuan Zhang, Yuchen Wang, Jianfei Yu
MIT / Peking University / Tsinghua University
AutoSurvey takes a research topic as input and automatically produces a structured, citation-rich literature survey. The system searches, reads, categorizes, and synthesizes papers to generate publication-quality reviews.
Self-Evolving Code: Autonomous Bug Fixing with Zero Human Intervention
Yuanjue Fan, Yujie Fei
Peking University
A self-evolving software system that continuously monitors production logs, identifies bugs, generates patches, validates them through automated testing, and deploys fixes — all without human involvement.
CLI Test: Automated Hyperparameter Tuning via LLM
Test Author A, Test Author B
Test University
A test work submitted via CLI to demonstrate the automated submission pipeline.