Hi, I am Xiaoxiao (Sia) Li. I will join UBC ECE as an assistant professor in Fall 2021. Now I am a postdoc in the Department of Computer Science at Princeton, working with Prof. Kai Li and Prof. Olga Troyanskaya. In 2020 summer, I obtained my Ph.D. degree in Biomedical Engineering from Yale University, where I was a member in Image Processing and Analysis Group(IPAG). I was fortunate to be advised by Prof. James Duncan. I received Yale Advanced Graduate Leadership Fellowship in 2018. I obtained my B.S. (honors degree) from Chu Kochen College, Zhejiang University, China, in June 2015.
My current research lies in the interdisciplinary field of deep learning, model explainability, data privacy, multi-site learning, medical image analysis, AI-based drug discovery and bioinformatics. I aim to tackle the ongoing challenges of data isolation, data privacy, and needs for model transparency in applying AI to healthcare applications. My research plan is to develop next-generation trustworthy AI systems.
If you want to work with me, please email me (click here) including your CV, transcript, and one of your research papers if applicable. Due to the large amount of emails I receive, I may not be able to respond to each one individually. To help me notice you email, please put "[ILOLLEH]", written backwards, in your email subject. All formats of local and remote collobration are welcomed.
News
[06/2021] One paper accepted at MICCAI 2021.
[05/2021] One paper accepted at ICML 2021.
[04/2021] We are organizing MICCAI 2021 workshop on Distributed and Collaborative Learning. Please submit your paper here.
[04/2021] We are organizing ICML2021 workshop on Interpretable ML in Healthcare. Please submit your paper here.
[01/2021] Proposal about drug discrovery funded by MSR & Mila Research ($65,000 CAD).
[01/2021] Two papers accepted at ICLR 2021.
[10/2020] Best Paper Award in DART 2020.
[08/2020] Awarded MICCAI Student Participation Award.
[07/2020] One paper accepted at Medical Image Analysis (IF 11.15).
[06/2020] One paper accepted at ICML 2020.
[05/2020] Two papers accepted at MICCAI 2020.
[03/2020] Passed Thesis Defense at Yale.
[10/2019] Best Paper Award in MIML 2019.
[06/2019] Started my internship at J.P. Morgan AI Research.
[06/2019] Two papers accepted at MICCAI 2019.
Recent Publications
For the complet list, please check my google scholar. Email me if you have any qeustions about my papers or code, or if you would like to collaborate with me.
FL-NTK: A Neural Tangent Kernel-based Framework for Federated Learning Analysis.
Huang, B., Li, X., Song, Z., & Yang, X.
ICML 2021.
[paper]
FedBN: Federated Learning on Non-IID Features via Local Batch Normalization.
Li, X., Jiang, M., Zhang, X., Kamp, M., & Dou, Q.
ICLR 2021.
[paper]
[code]
On InstaHide, Phase Retrieval, and Sparse Matrix Factorization.
Chen, S., Li, X., Song, Z., & Zhuo, D.
ICLR 2021.
[paper]
Multi-site fMRI Analysis Using Privacy-preserving Federated Learning and Domain Adaptation: ABIDE Results.
Li, X., Gu, Y., Dvornek, N., Staib, L., Ventola, P., & Duncan, J. S.
Medical Image Analysis.
[paper]
[arxiv]
[code]
Efficient Shapley Explanation For Features Importance Estimation Under Uncertainty.
Li, X., Zhou, Y., Dvornek, N., Gu, Y., Ventola, P., & Duncan, J. S.
MICCAI 2020.
[paper]
[code]
Pooling Regularized Graph Neural Network for fMRI Biomarker Analysis.
Li, X., Zhou, Y., Dvornek, N., Zhang, M., Zhuang, J, Ventola, P., & Duncan, J. S.
MICCAI 2020.
[paper]
[arxiv]
[code]
Adaptive Checkpoint Adjoint Method for Gradient Estimation in Neural ODE.
Zhuang, J., Dvornek, N.C., Li, X., Tatikonda, S., Papademetris, S.,and Duncan, J.S.
ICML 2020.
[paper]
[arxiv]
[code]
Efficient Interpretation of Deep Learning Models Using Graph Structure and Cooperative Game Theory: Application to ASD Biomarker Discovery.
Li, X., Dvornek, N.C., Zhou, Y., Zhuang, J., Ventola, P. and Duncan, J.S.
IPMI 2019.
[paper]
[arxiv]
Interpretable Multimodality Embedding of Cerebral Cortex Using Attention Graph Network for Identifying Bipolar Disorder.
Yang, H.*, Li, X.*, , Wu, Y., Li, S., Lu, S., Duncan, J.S., Gee, J.C. and Gu, S.
MICCAI 2019.
[paper]
[arxiv]
Graph Neural Network for Interpreting Task-fMRI Biomarkers.
Li, X., Dvornek, N.C., Zhou, Y., Zhuang, J., Ventola, P. and Duncan, J.S.
MICCAI 2019.
[paper]
[arxiv]
Recent Projects
Here are my recent exciting projects. For the full list of my projects, please refer to my publications.
Federated Learning on Non-iid Features via Local Batch Normalization
Most of the previous federated learning work has focused on a difference in the distribution of labels. Unlike those settings, we address an important problem of FL, e.g., different scanner/sensors in medical imaging, where local clients may store examples with different marginal or conditional feature distributions compared to other nodes, which we denote as feature shift non-iid. In this work, we propose an effective method that uses local batch normalization to alleviate the feature shift before averaging models.
Transfer Pruning for Inversion Attack Defense
We propose to prune a target neural network with a public datasetand transfers it to the distributed learning setting to collaboratively train the pruned network with distributed private datasets via sharing a hidden layer of representations, without sharing the data. Our experiments show that the proposed approach achieves a much better trade-off between accuracyand privacy preservation.
Privacy-preserving Multi-site Medical Imagse Analysis
We present a privacy-preserving federated learning framework for multi-site fMRI analysis. To overcome the domain shift issue, we have proposed two strategies: MoE and adversarial domain alignment to boost federated learning model performance. We have shown federated learning performance can potentially be boosted by adding domain adaptation and discussed the condition of benefits. Our approach brings new hope for accelerating deep learning applications in the field of medical imaging, where data isolation and the emphasis on data privacy have become challenges.
BrainGNN: Explainable Graph Neural Network for Neuroimaging Analysis
BrainGNN takes graphs built from neuroimages as inputs, and then outputs prediction results together with interpretation results.With the built-in interpretability, BrainGNN not only performs better on prediction than alternative methods, but also detects salient brain regions associated with predictions and discovers brain community patterns.
DistDeepSHAP: Efficient Shapley Explanation for Feature Importance Estimation Under Uncertainty
We propose DistDeepSHAP, a post-hoc feature importance estimation method under uncertainty evaluation for deep learning models.First, it can obtain uncertainty estimates for the provided feature importance scores. Second, it can better utilize the empirical distribution and has the potential for better feature importance scores estimation from the generated percentile sliency maps. Last but not least, it can compare with arbitrary subgroup references and interpret subgroup salient features, which is crucial for medical image study.
Talks
[02/2021] Invidted talk about Trustworthy AI for Healthcare at Emory University.
[01/2021] Invidted talk about Trustworthy AI for Healthcare at Istanbul Technical University.
[11/2020] Invidted talk about BrainGNN at Tulune University.
[11/2020] Invidted talk about BrainGNN at Lehigh University.
[09/2020] Invidted talk about Multi-site Federated Learning for NeuroImaging Anslysis at Chinese Academy of Sciences.
Teaching
[2017, 2018] ENAS 194: Ordinary and Partial Differential Equation (Yale University)
[2017, 2019] BENG 352: Biosignal Processing (Yale University)
[2021] COS 598D: Systems and Machine Learning (Princeton University) [Course website]
[2022] ELEC 400M: Machine Learning Fundamentals for Engineers (UBC) [Syllabus]
Students
Chun-Yin Huang (PhD of 2021, from Carneigie Mellon University)
Sana Ayromlou (MASc of 2021, from Sharif University of Technology, co-advised with Prof. Purang Abolmaesumi)
