Research interests

My research lies in the broad area of Statistics, Data Science and Machine Learning, with a special interest in developing novel and principled methods to handle real data with spatial-temporal dependence. In particular, much of my work focuses on answering “what if” counterfactual questions and elucidating causal connections within the context of healthcare; The ultimate objective of my work is to empower healthcare professionals with enhanced situational awareness, promote individualized treatment, and potentially steer biomedical research to validate the identified causal relationships. 

Specifically, I work on:

1. Trustworthy AI: causality (causal discovery, causal inference, and difference in difference) and fairness,

2. Reinforcement learning for generative AI (e.g., RLHF),

3. Time series and point process,

4. Hypothesis testing (two-sample test, goodness-of-fit test, and change-point detection),

and their applications in healthcare, power systems, finance, psychology, criminology, and so on.

Journal articles

1. Causal Graph Discovery from Self and Mutually Exciting Time Series. [arXiv]
   Song Wei, Yao Xie, Christopher S. Josef, Rishikesan Kamaleswaran.
   IEEE Journal on Selected Areas in Information Theory, vol. 4, pp. 747-761, 2023.
   Preliminary version presented at ICML 2021 Workshop on Interpretable Machine Learning in Healthcare (IMLH) [link]

2. Online Kernel CUSUM for Change-Point Detection. [arXiv][code]
   Submitted to Journal of the Royal Statistical Society: Series B
   Song Wei, Yao Xie.

3. Transfer Learning for Causal Effect Estimation. [arXiv][code]
   Song Wei, Hanyu Zhang, Ronald Moore, Rishikesan Kamaleswaran, Yao Xie.
   Submitted to Springer Book on Big Data Analysis, Biostatistics and Bioinformatics  

Conference proceedings

1. Assessing Electricity Service Unfairness with Transfer Counterfactual Learning. [arXiv]
   Song Wei, Xiangrui Kong, Alinson Santos Xavier, Shixiang Zhu, Yao Xie, Feng Qiu.
   Submitted.

2. Granger Causal Chain Discovery for Sepsis-Associated Derangements via Continuous-Time Hawkes Processes. [arXiv][code]
   Song Wei, Yao Xie, Christopher S. Josef, Rishikesan Kamaleswaran.
   ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD), 2023

3. Causal Structural Learning from Time Series: A Convex Optimization Approach. [arXiv]
   Song Wei, Yao Xie.
   Asilomar Conference on Signals, Systems, and Computers, 2023
   Preliminary version presented at IJCAI 2023 Workshop on AI for Time Series Analysis (AI4TS)

4. Inferring serial correlation with dynamic backgrounds. [arXiv]
   Song Wei, Yao Xie, Dobromir Rahnev.
   International Conference on Machine Learning (ICML), 2021 (Long presentation, acceptance rate: 166/5513 = 3%)

5. Goodness-of-Fit Test of Mismatched Models for Self-Exciting Processes. [arXiv]
   Song Wei, Shixiang Zhu, Minghe Zhang, Yao Xie.
   International Conference on Artificial Intelligence and Statistics (AISTATS), 2021 

6. Noisy Gradient Descent Converges to Flat Minima for Nonconvex Matrix Factorization. [arXiv]
   Tianyi Liu, Yan Li, Song Wei, Enlu Zhou, Tuo Zhao.
   International Conference on Artificial Intelligence and Statistics (AISTATS), 2021 

Workshop papers

1. INTAGS: Interactive Agent-Guided Simulation. [Paper]
   Song Wei, Andrea Coletta, Svitlana Vyetrenko, Tucker Balch.
   Workshop on Synthetic Data Generation with Generative AI (SyntheticData4ML), NeurIPS, 2023 [link]

2. Unfairness Detection within Power Systems through Transfer Counterfactual Learning.
   Song Wei, Xiangrui Kong, Sarah Ann Huestis-Mitchell, Shixiang Zhu, Yao Xie, Alinson Santos Xavier, Feng Qiu.
   Workshops on Causal Representation Learning (CRL) [link] and Algorithmic Fairness through the Lens of Time (AFT) [link], NeurIPS, 2023 

3. Transfer Causal Learning: Causal Effect Estimation with Knowledge Transfer.
   Song Wei, Ronald Moore, Hanyu Zhang, Yao Xie, Rishikesan Kamaleswaran.
   3rd Workshop on Interpretable Machine Learning in Healthcare (IMLH), ICML, 2023 [link]

4. Entropic Regularization for Adversarial Robust Learning
   Jie Wang, Yifan Lin, Song Wei, Rui Gao, Yao Xie.
   18th INFORMS DMDA Workshop (Winner, Best Paper Competition, 2 out of 57) [link]

Technical report

• Optimal Sub-sampling to Boost Power of Kernel Sequential Change-point Detection. [arXiv]
  Song Wei, Chaofan Huang.