Qiuhao Zeng
Postdoctoral Fellow · University of Toronto
About
I am a postdoctoral fellow at the University of Toronto, supervised by Prof. Rahul Krishnan. Previously, I completed my PhD at the University of Western Ontario, where I had a wonderful four years working with Prof. Boyu Wang and Prof. Charles Ling (Fellow, Canadian Academy of Engineering). My research spans efficient Transformer architectures, long-context data modeling (including tabular and language data), transfer learning under temporal distribution shifts, graph learning, causal inference, and AI for science.
Research Interests
- Long-context efficient attention and Transformer acceleration (Triton kernels, top-k/sparse attention).
- Causal inference and AI-for-science applications.
- Temporal distribution shift & domain generalization for sequence modeling (evolving time-series and latent trajectories).
Publications
-
Graph Domain Adaptation via Homophily-Agnostic Reconstructing StructureAAAI Conference on Artificial Intelligence (AAAI), 2026 (Oral).
-
Investigating the Multilingual Calibration Effects of Language Model Instruction TuningThe 19th Conference of the European Chapter of the Association for Computational Linguistics (EACL), 2026 (Main Conference).
-
Entropy-Guided Dynamic Tokens for Graph-LLM Alignment in Molecular UnderstandingInternational Conference on Learning Representations (ICLR), 2026 (Poster).
-
ZETA: Leveraging Z-order Curves for Efficient Top-k AttentionInternational Conference on Learning Representations (ICLR), 2025.
-
Mamba Modulation: On the Length Generalization of Mamba ModelsConference on Neural Information Processing Systems (NeurIPS), 2025.
-
Calibrated Language Models and How to Find Them with Label SmoothingInternational Conference on Machine Learning (ICML), 2025.
-
On the Benefits of Attribute-Driven Graph Domain AdaptationInternational Conference on Learning Representations (ICLR), 2025.
-
Homophily Enhanced Graph Domain AdaptationInternational Conference on Machine Learning (ICML), 2025.
-
Towards Understanding Evolving Patterns in Sequential DataConference on Neural Information Processing Systems (NeurIPS), 2024 (Spotlight; Top 2.1%).
-
Latent Trajectory Learning for Limited Timestamps under Distribution Shift over TimeInternational Conference on Learning Representations (ICLR), 2024 (Oral; Top 1.2%).
-
Foresee What You Will Learn: Data Augmentation for Domain Generalization in Non-Stationary EnvironmentsAAAI Conference on Artificial Intelligence (AAAI), 2023.
-
LGGNet: Learning from Local-Global-Graph Representations for Brain-Computer InterfaceIEEE Transactions on Neural Networks and Learning Systems (TNNLS), 2023.
Academic Service
- Conference Reviewer: ICLR, NeurIPS (Top Reviewer), ICML, AAAI, AISTATS.
- Journal Reviewer: TNNLS, TMLR.
- Teaching Assistant: CS3346 (AI), CS2210 (Data Structures), CS3357 (Networks).
Patent
Mental Arousal Level Regulation System and Method, PCT Patent no. PCT/SG2022/050243 (2022).
Contact
Email: qiuhao.zeng@utoronto.ca
Also on LinkedIn and Google Scholar.