崔鹏 - 稳定学习-融合-30分钟.pdf

《崔鹏 - 稳定学习-融合-30分钟.pdf》由会员分享，可在线阅读，更多相关《崔鹏 - 稳定学习-融合-30分钟.pdf（35页珍藏版）》请在三个皮匠报告上搜索。

1、Peng Cui Tsinghua University Stable Learning: The Convergence of Machine Learning and Causal Inference 2 Human Healthcare Law Transportation Fintech Now AI is stepping into risk-sensitive areas Shifting from Performance Driven to Risk Sensitive Human in the loopUnexplainable MedicalMilitaryFinance R

2、isks of Todays AI Algorithms 4 Most ML methods are developed under I.I.D hypothesis Risks of Todays AI Algorithms 5 Yes Maybe No Risks of Todays AI Algorithms Cancer survival rate prediction 6 Training Data Predictive Model Testing Data City Hospital University Hospital Features: Body status Income

3、Treatments Medications Higher income, higher survival rate. City Hospital Survival rate is not so correlated with income. Risks of Todays AI Algorithms 7 The Current Condition ExplainabilityStability We cannot understand AIWe dont trust AI Dilemma 8 A plausible reason: Correlation Correlation is the very basics of machine learning. 9 Correlation is not explainable 10 Correlation is unstable 11 Its