Tenure Talk
Causality at Machine Learning Scale and Complexity
Distinguishing coincidence from cause-and-effect relationships is a central challenge in AI (and science): which patterns just co-occur, and which signal causality? Statistical causality offers a rigorous framework for this question, but it assumes clean, low-dimensional, structured data, which clashes with the raw, high-dimensional observations of machine learning applications, including in science. In this talk, I will present how we can now learn causal structure directly from unstructured data using deep learning, bridging between the modeling power of causality and the scalability properties of machine learning. I will also highlight some cross-disciplinary collaborations my group has built, uniquely enabled by ISTA.