Zijian Li (李梓健)
Postdoctoral Researcher
About
Hi! I’m Zijian Li, a postdoctoral researcher at MBZUAI/Carnegie Mellon University, advised by Prof. Kun Zhang. Before that, I received my Ph.D. from Guangdong University of Technology, where I was advised by Prof. Ruichu Cai. I have had the privilege of interning at YITU Technology, WeChat, and the Advanced Digital Sciences Center (ADSC) in Singapore.
My research interests include causal discovery and its applications， including time-series analysis, and transfer learning. My long-term research goal is to build reliable AI systems with causal-thinking capabilities—systems that go beyond correlation-based pattern recognition to uncover, reason about, and leverage the underlying causal mechanisms of complex environments. By integrating causal discovery, temporal modeling, and transfer learning, I aim to develop AI systems that are robust under distribution shifts, interpretable in their decisions, and capable of reliable generalization across domains, tasks, and changing environments.
News
- 2 papers on causal discovery and causal representation learning were accepted by ICML 2026.
- 1 paper on time-series forecasting was accepted by IJCAI 2026.
- 1 paper on causal representation learning was accepted by CVPR 2026.
- 3 papers on causal representation learning were accepted by ICLR 2026.
- 4 papers were accepted by NeurIPS 2026.
Services
Professional Service
- Area Chair: for ICLR 2026, NeurIPS 2026.
- Conference Reviewer / Program Committee Member: ICML, CVPR, IJCAI and so on.
- Journal Reviewer: TPAMI, JMLR, TNNLS, and so on.
Selected Publications
All papers →
Learning General Causal Structures with Hidden Dynamic Process for Climate Analysis
Diverse Dictionary Learning
Online Time Series Forecasting with Theoretical Guarantees
Towards Identifiability of Hierarchical Temporal Causal Representation Learning
Identifying Semantic Component for Robust Molecular Property Prediction
Causal View of Time Series Imputation: Some Identification Results on Missing Mechanism
From Orthogonality to Dependency: Learning Disentangled Representation for Multi-Modal Time-Series Sensing Signals
Synergy Between Sufficient Changes and Sparse Mixing Procedure for Disentangled Representation Learning
Time Series Domain Adaptation via Latent Invariant Causal Mechanism
Disentangling Long-Short Term State Under Unknown Interventions for Online Time Series Forecasting
From Orthogonality to Dependency: Learning Disentangled Representation for Multi-Modal Time-Series Sensing Signals
On the Identification of Temporally Causal Representation with Instantaneous Dependence
Causal Temporal Representation Learning with Nonstationary Sparse Transition
