Speaker: Dr. Kun ZHANG

Deep networks meet causal modeling


Deep learning has been successful across a variety of machine learning tasks, including generating powerful features from raw input. It, however, usually requires a large number of training data that cover all possible scenarios to be considered. On the other hand, we care about causal models in our daily life, scientific discovery, and decision making; they represent intuitive, stable relationships in complex systems and enable making predictions under interventions or data heterogeneity. We aim to combine deep networks and causal representations for improved transfer learning and concept learning. In particular, I will show how deep networks, combined with latent variable modeling, can be used in nonparametric transfer learning for understanding and adapting to new scenarios. I will also discuss how properties of causal representations help learn latent causal features with deep networks.


About Kun Zhang

Kun Zhang is an assistant professor in the philosophy department and an affiliate faculty in the machine learning department of Carnegie Mellon University, USA. He is also a senior research scientist at Max-Planck Institute for Intelligent Systems, Tübingen, Germany. His research interests lie in machine learning and artificial intelligence, especially in causal discovery and causality-related learning. He has made a series of contributions in solving some long-standing problems in causality, such as how to distinguish cause from effect, how to make nonparametric conditional independence test reliable, and how causality facilitates underlying and solving machine learning problems. He won the Best Benchmark Award of the 2008 causality challenge, and has served as air chairs or senior program committee members for a number of conferences in machine learning or artificial intelligence, including NIPS, UAI, ICML, and IJCAI, and as an associate editor for three journals including Pattern Recognition. He has organized various academic activities to foster interdisciplinary research in causality.