Shao Pengyang

Shao Pengyang

Strike while the iron is hot

About

I am Shao Pengyang (邵鹏阳), currently a postdoctoral researcher at the National University of Singapore, mentored by Prof. Tat-Seng Chua and Yunshan Ma. I received my Ph.D. in July 2025 from Hefei University of Technology, under the supervision of Prof. Meng Wang. My current research focuses on AI+X and trustworthy AI, with the goal of developing reliable and responsible AI techniques and applying them to meaningful real-world problems across different domains.

I have also mentored students on several research projects:

  • Naixin Zhai, whose first-author paper was accepted by the ACL 2026 Main Conference and received an Outstanding Paper Award;
  • Jilong Liu, whose first-author paper was accepted by AAAI 2026 as an Oral Presentation;
  • Chao Chen, a Ph.D. candidate at Hefei University of Technology, on two projects concerning LLM personalization and over-refusal in LLM safety alignment;
  • Yanzheng Jin, on two projects concerning LLM unlearning and AI for finance;
  • Chuanpeng Lu, on an ongoing project developing a new LLM unlearning benchmark.

I am also happy to connect and recommend qualified students for relevant industry internship opportunities. If you are interested in collaboration, please feel free to contact me.

I am currently exploring research positions related to AI+X and trustworthy AI. I am enthusiastic about expanding into new AI application domains beyond my current research areas, rather than limiting myself to a fixed set of applications. Regardless of the specific domain, my central research interest remains the same: understanding whether AI systems can be developed, evaluated, and deployed in a trustworthy, reliable, safe, and responsible manner.

If you are interested in my research, potential collaborations, or relevant opportunities, please feel free to contact me.

Email: shaopymark at gmail.com · shaopymark at nus.edu.sg

Featured Publications

acl.png
Maximizing Local Entropy Where It Matters: Prefix-Aware Localized LLM Unlearning. ACL 2026 Oral Presentation. [pdf] [code]
We propose PALU (Prefix-Aware Localized Unlearning), a framework driven by local entropy maximization across both temporal and vocabulary dimensions. PALU shows that suppressing the sensitive prefix alone is sufficient to sever the causal generation link, and flattening only the top-K logits is adequate to maximize uncertainty in the critical subspace.
paper figure
BalDRO: A Distributionally Robust Optimization based Framework for LLM Unlearning. WWW 2026 Oral Presentation. [pdf] [code]
We propose BalDRO, a distributionally robust optimization framework that addresses sample-wise imbalance in LLM unlearning. By formulating unlearning as a min–sup optimization, BalDRO adaptively emphasizes hard-to-forget samples and prevents over-forgetting of easy ones.

Important Publications

* denotes corresponding author. † denotes equal contribution.

  • Naixin Zhai, Pengyang Shao*, Binbin Zheng, Yonghui Yang, Fei Shen, Long Bai, Xun Yang*. Maximizing Local Entropy Where It Matters: Prefix-Aware Localized LLM Unlearning. ACL 2026. [pdf]
  • Pengyang Shao, Le Wu, Kun Zhang, Lei Chen, Meng Wang. Privacy Matters: Data Attack to Make User Preferences Unlearnable in Recommendation. ACM TOIS 2026. [pdf]
  • Pengyang Shao, Naixin Zhai, Lei Chen, Yonghui Yang, Fengbin Zhu*, Xun Yang*, Meng Wang. BalDRO: A Distributionally Robust Optimization based Framework for Large Language Model Unlearning. WWW 2026. [pdf]
  • Pengyang Shao, Lei Chen*, Fei Liu, Yonghui Yang, Xun Yang, Meng Wang*. Multi-Agent Debate based Concept Augmentation for Enhanced Cognitive Diagnosis. KDD 2026 August Cycle. [pdf]
  • Jilong Liu, Pengyang Shao*, Wei Qin, Fei Liu, Yonghui Yang, Richang Hong*. Debate over Mixed-knowledge: A Robust Multi-Agent Framework for Incomplete Knowledge Graph Question Answering. AAAI 2026. [pdf]
  • Pengyang Shao†, Yonghui Yang†, Chen Gao, Lei Chen, Kun Zhang, Chenyi Zhuang, Le Wu, Yong Li, Meng Wang*. Exploring Heterogeneity and Uncertainty for Graph-based Cognitive Diagnosis Models in Intelligent Education. KDD 2025. [pdf]
  • Pengyang Shao, Le Wu*, Kun Zhang, Defu Lian, Richang Hong, Yong Li, Meng Wang*. Average User-side Counterfactual Fairness for Collaborative Filtering. ACM TOIS, 2024. [pdf]
  • Pengyang Shao, Le Wu*, Lei Chen, Kun Zhang, Meng Wang*. FairCF: Fairness-aware collaborative filtering. Science China Information Sciences, 2022. [pdf]
  • Pengyang Shao, Zihan Wang, Junsong Xie, Yonghui Yang, Meng Wang. Towards Reliable Cross-Domain Recommendation: A Disentangled Global Graph Learning based Framework. Frontiers of Computer Science, 2025. [pdf]
  • Pengyang Shao, Kun Zhang*, Chen Gao*, Lei Chen, Miaomiao Cai, Le Wu, Yong Li, Meng Wang. Breaking student-concept sparsity barrier for cognitive diagnosis. Frontiers of Computer Science, 2025. [pdf]

Academic Services

  • Journal Reviewer: IEEE TKDE/TBD, ACM ToRS/TOIS/TiiS, npj Health Systems, Pattern Recognition, etc.
  • Conference Reviewer: SIGIR, KDD, WWW, AAAI, ACL, EMNLP, CIKM, etc.

Awards

  • 2026.07 Outstanding Paper Award at ACL 2026
  • 2025.05 Best Paper Award at TIME 2025 (Workshop on TheWebConf 2025)
  • 2024.12 China Association for Science and Technology (CAST) Young Talent Support Program (PhD Special Track)
  • 2021.09 First-Class Graduate Academic Scholarship (Awarded Annually, 2021–2024)
  • 2019.06 Outstanding Graduate of Anhui Province