Towards Trustworthy AI: From High-dimensional Statistics to Causality
Tuesday, November 23, 2021
Although current AI models have achieved remarkable performance in various fields,they suffer from trustworthy-related issues (e.g., explainability, reproducibility, robustness and fairness), which are key bottlenecks towards real applications. To address this problem, we depart from current performance-driven paradigm by resorting to statistical learning with theoretical guarantees from the perspective of inference. In this talk, Dr. Xinwei Sun will introduce our series of works, that pave a road towards trustworthy AI models, from high-dimensional statistics to causality. Specifically, he will first introduce a differential-inclusion based method for sparse recovery together with its false-discovery rate (FDR) analysis; followed by causal inference tools (intervention, counterfactual) towards robustness and explainability. Theoretical analysis regarding consistency and identifiability will be introduced. Empirically, I will show the utility of these works on medical imaging analysis.
Speaker/s
Dr. Xinwei Sun is now a researcher in the machine learning group at Microsoft Research Asia (MSRA). Prior to joining MSRA, Dr. Sun obtained BS degree in pure mathematics from Nankai university in 2013, and PhD degree in statistics from Peking University in 2018. His research interests focus on the statistical models (such as high-dimensional statistics, causal inference, etc.) and their applications on medical imaging analysis. He has published more than 20 papers in top mathematical journals such as Applied Computational Harmonic Analysis (AHCA), machine learning conferences such as NeurIPS, ICML, and medical imaging-related conferences such as MICCAI, CVPR, ICCV, ECCV, etc. He has served as (senior) program committee member in NeurIPS, ICML, ICLR, AAAI, IJCAI, CVPR. Many of his works on medical imaging analysis have been transformed into medical products and applied in renowned hospitals in China.