Hello, I’m
Siyu Wang
Postdoctoral Fellow UNSW Sydney
How can we build machine learning systems that reason about cause and effect and make reliable decisions?
I study causal inference, representation learning and generative models, with applications in reinforcement learning and recommender systems.
Click or tap the photo to pause or resume decorative motion throughout the page.
News
-
NeurIPS Paper accepted Main Track
When Do Learned State Representations Break Sensitivity Analysis?
-
ICML Two papers accepted Main Track
-
SIGIR Paper accepted Full Paper
Advantage-Conditioned Flow Policy for Offline Reinforcement Learning in Recommendation
-
The Web Conference Paper published Main Track · Full Paper
Policy-Guided Causal State Representation for Offline Reinforcement Learning Recommendation
Selected research
-
Reward-Preserving Counterfactual State Editing for Offline Reinforcement Learning
Offline policies can rely on misleading correlations. We edit recorded states while preserving predicted rewards, helping policies focus on causally relevant information.
Read paper -
Continuity-Regularized Flow Matching for Offline Reinforcement Learning
Generative policies can follow unstable paths from noise to actions. We regularise these paths to make offline policy learning smoother and more stable.
Read paper -
Policy-Guided Causal State Representation for Offline Reinforcement Learning Recommendation
Recommendation logs contain irrelevant and noisy features. We learn compact state representations that retain information causally relevant to reward.
Read paper
Research interests
- Causal inference & representation learning
How can we learn representations that capture causal relationships and support reliable decisions?
Related paper: Counterfactual state editing → - Generative models & reinforcement learning
How can generative models help us learn effective decision-making policies from existing data?
Related paper: Flow-based policy learning → - Reliable machine learning & evaluation
When can we trust a model’s conclusions, and how can we identify where its assumptions break down?
Related paper: Learned representations and sensitivity analysis ↗
About me
I’m a Postdoctoral Fellow at UNSW Sydney, supervised by Prof. Lina Yao, who also supervised my PhD in Computer Science and Engineering at UNSW. Before joining UNSW in my current role, I worked at Macquarie University under the supervision of Prof. Michael Sheng.
I enjoy the collaborative side of research: turning an initial question into a clear problem, testing ideas carefully, and understanding what the results tell us. When mentoring research students, I aim to help them develop independent thinking, thoughtful experiments and clear scientific communication.
Academic service
- Senior programme committee
- IEEE ICDH 2026
- Programme committee member
- The Web Conference 2024–2025
- CIKM 2024–2025
- KDD 2025–2026
- ACL ARR 2026
- ACM MM 2026
- NeurIPS 2026
- ICLR 2026
- Journal reviewer
- ACM TOIS
- IEEE TKDE
- IEEE TNNLS
- World Wide Web
- ACM TOSN
- International Journal of Data Science and Analytics
- ACM TALLIP
Contact
I welcome opportunities to collaborate and explore new research ideas. If your interests overlap with mine, I’d love to hear from you.
