I am a Ph.D. candidate in Systems Engineering at Beihang University, advised by Prof. Shihai Wang. My research lies at the intersection of software engineering and trustworthy artificial intelligence, with a particular focus on software engineering for AI (SE4AI). Specifically, I am interested in the reliability, testing, and repair of deep neural networks. My goal is to improve the robustness and deployment reliability of AI systems through principled testing prioritization and model repair techniques.

🔥 News

  • 2026.06:  🎉 Our paper on When Ambiguity Meets Atypicality: Dual-Perspective Test Input Prioritization for DNNs (DuFP) has been directly accepted by ASE 2026 (115/1,236 reviewed submissions, top 9.3%)!
  • 2025.09:  🎉 Our paper on SMiR: Minimizing Side-Effects in Repairing Specific Misclassifications for Deep Neural Networks has been accepted by TOSEM! This is the first TOSEM paper in our group!

📝 Publications

Representative Works

TOSEM
sym

SMiR: Minimizing Side-Effects in Repairing Specific Misclassifications for Deep Neural Networks

Haoran Li, Shihai Wang, Bin Liu, Tengfei Shi, Wentao Wu, Yu Liu

TOSEM (CCF-A, JCR-Q1)

Paper | Code SMiR GitHub stars

  • We propose SMiR, an effective repair framework for addressing specific misclassifications with marginal accuracy loss. Rather than patching the original model, our method localizes and recovers the distorted feature dimensions to mitigate their negative impact on the distinction between specific classes. Moreover, SMiR narrows the scope of repair to the filtered suspect set, protecting other well-trained decision boundaries from disturbance. On average, our approach repairs 27%-36.2% more specific misclassifications than the baselines, while introducing only 7.8%-70.9% as many new errors. Besides, SMiR maintains state-of-the-art performance on safety-critical datasets.
ASE-2026
sym

When Ambiguity Meets Atypicality: Dual-Perspective Test Input Prioritization for DNNs

Haoran Li, Shihai Wang, Bin Liu, Jialuo Chen, Wenjing Zhu, Yu Liu, Tengfei Shi, Shudi Guo;

ASE-2026 (CCF-A)

Paper | Code DuFP GitHub stars

  • We propose DuFP (Dual perspective Feature space Prioritization), a KNN density-based test input prioritization approach for DNNs that jointly incorporates inter-class and intra-class perspectives. Built upon class-conditional density estimation, DuFP characterizes prediction correctness through an ambiguity score and an atypicality score, then integrates them into a hybrid uncertainty score to guide prioritization. Extensive evaluations on image and text datasets under clean, corrupted, and adversarial scenarios show that DuFP effectively and efficiently prioritizes failure-inducing inputs and outperforms state-of-the-art approaches.

All Publications

🎖 Honors and Awards

  • 2020.12 National Scholarship (Top 1%)
  • 2020.05 Zhangjiagang Municipal Government Scholarship (Top 2%)

📖 Educations

  • 2022.09 – now, Ph.D. Student, Beihang University, School of Reliability and Systems Engineering (BUAA: China, Beijing)
  • 2021.09 – 2022.08, Master, Beihang University, School of Reliability and Systems Engineering (BUAA: China, Beijing)
  • 2017.09 – 2021.07, Undergraduate (1/58), Nanjing Tech University (NJTech: China, Nanjing)

💬 Talks & Presentations

  • ASE 2026

    Presented "When Ambiguity Meets Atypicality: Dual-Perspective Test Input Prioritization for DNNs" at ASE 2026 in Munich, Germany, and participated in the poster presentation.

    DuFP poster presented at ASE 2026
  • QRS 2022

    Presented "TSDTest: An Efficient Coverage Guided Two-Stage Testing for Deep Learning Systems" at QRS 2022 online.

💻 Internships

  • 2026.07–2026.08, Huawei - AI Application Algorithm Engineer, 2012 Laboratories, China.

Open-Source Projects

  • A curated reading list for deep neural network testing and repair, covering coverage metrics, test generation, test selection and prioritization, and model repair. Includes surveys and empirical studies, with publication details and links to available papers and code.

    Browse the DNN paper collection: website preview in light and dark themes