Sun Changsheng

About

I am a final-year Ph.D. candidate at the PLSE Lab, School of Computing, National University of Singapore, advised by Prof. Dong Jin Song. My research focuses on building trustworthy and reliable AI systems—making modern ML models robust, interpretable, and auditable.

🎓 I am expected to graduate in early 2026 and am actively looking for job opportunities.

Research Interests

• AI Safety & Robustness: Adversarial attacks and defenses, risk-aware optimization, with applications to GNNs, point clouds, and LLMs.
• Interpretability & Explainability: Explanation methods for model decisions, directionality-aware and risk-aware explanations, model debugging and auditing.
• Geometric Deep Learning: Graph neural networks, point clouds, structure-aware representation learning.

Education

• Doctor of Philosophy in Computer Science (2022 - Mar 2026)
  National University of Singapore
  Thesis: Trustworthy Geometric Learning: From Structural Biases to Risk-Aware Robustness
  Advisor: Prof. Dong Jin Song
• Master of Computing in Artificial Intelligence (2020 - 2021)
  National University of Singapore
• Bachelor of Engineering in Computer Science (2015 - 2019)
  Xidian University, Xi'an, China

Work Experience

• Research Assistant (Jan 2026 - Present)
  NUS School of Computing
  Working on AISG project: Trustworthy Geometric Learning
  Host: Prof. Dong Jin Song and Dr. Zhang Yedi
• Research Assistant (July 2021 - Dec 2021)
  NUS School of Computing
  Built and maintained a reproducible evaluation pipeline for trustworthiness assessment of deep learning systems, emphasizing uncertainty-aware reliability analysis. Contributed to work published at ASE 2022 (InputReflector).
  Host: Prof. Dong Jin Song and Prof. Xiao Yan
• Research Intern (Jan 2020 - June 2020)
  NUS-Singtel Cyber Security R&D Lab
  Developed a white-box testing approach for deep neural networks by leveraging intermediate-layer signals and density estimation for reliability assessment. Contributed to work published at ICSE 2021 (Self-Checking Deep Neural Networks).
  Host: Prof. David S. Rosenblum
• Research Intern & Algorithm Engineer (Jan 2018 - June 2018)
  JD Intelligent Cities Research, JD.com (MSRA Urban Computing Group)
  Prototyped a distributed DQN research system on spatiotemporal data using the Ray execution framework; focused on training scalability and system performance profiling.
  Host: Prof. Yu Zheng and Dr. Junbo Zhang

Skills

Programming & Scripting: Python, Bash/Shell Scripting, SQL
AI & ML: PyTorch, TensorFlow, Hugging Face Transformers, Geometric Deep Learning (PyG, DGL), LLM Safety & Alignment
Systems: Linux/Unix, Multi-GPU setups, Ray, Docker, Git, ClickHouse, Spark
Miscellaneous: Trustworthy AI (Robustness, Explainability), Graph Learning, LLM Safety, Agentic AI

Services & Awards

• Reviewer/Program Committee
  • NeurIPS, ICML, AAAI, WWW, FSE, ASE (2022-2026)
• Teaching Assistant
  • CS5232 Formal Specification and Design Techniques (2024)
  • CS4218 Software Testing (2023)
  • CS4211 Formal Methods for Software Engineering (2022)
• Awards and Grants
  • Graduate Student Travel Grant 2025: IEEE Symposium on Security and Privacy 2025
  • Graduate Student Travel Grant 2024: AAAI Conference on Artificial Intelligence 2024
  • Graduate Student Travel Grant 2022: 37th IEEE/ACM Automated Software Engineering (ASE)
  • National University of Singapore Research Scholarship 2022-2026

Publications