Events Calendar

AI Seminar: General Methods for Causal Discovery

Speaker: Frederick Eberhardt, Caltech, Professor of Philosophy at Caltech

Talk Title: General Methods for Causal Discovery

Series: AISeminar

Abstract: This talk will consist of two parts, both concerned with causal discovery. In the first part I will describe some methods we have developed to perform causal inference with very weak background assumptions. The problem of causal discovery is converted into a constraint optimization problem that is then solved using maxSAT solvers. This general framework allows for the inclusion of very general background knowledge and only requires extremely weak model space assumptions. We show in simulations that for the restricted domains in which an exact Bayesian computation can be performed, our methods achieve an accuracy very close to that of the exact Bayesian computation. For more general domains, no competing method exists. In the second part, I will describe some very recent research we have done on extracting causes of behavior from images. This research addresses the more general question of how causal macro-variables can be constructed from micro-variables (pixel data, in our case). We hope that our account can lay the groundwork for a more general approach to automated causal analysis of images and video data.

Here are the relevant papers:
First part: http://people.hss.caltech.edu/~fde/papers/HEJ_UAI2014.pdf
Second part: http://arxiv.org/abs/1412.2309




Biography: Frederick Eberhardt is Professor of Philosophy at Caltech. He is primarily interested in methods for causal discovery from statistical data, the use of experiments in causal discovery, the integration of causal inferences from different data sets and the philosophical issues at the foundations of causality and probability. His research focuses both on how methods of causal discovery can be constructed and on how humans and animals in fact learn about causal relations. Before joining Caltech in 2013, he was assistant professor in the philosophy-neuroscience-psychology program at Washington University in St Louis. He got his PhD from Carnegie Mellon in 2007 and was a postdoc in psychology at UC Berkeley until 2009.

Host: Kun Zhang

Webcast: http://webcasterms1.isi.edu/mediasite/Viewer/?peid=44b5327011024c03bc46967a538adc9d1d

Location: Information Science Institute (ISI) - 1135

WebCast Link: http://webcasterms1.isi.edu/mediasite/Viewer/?peid=44b5327011024c03bc46967a538adc9d1d

Audiences: Everyone Is Invited

Contact: Alma Nava / Information Sciences Institute