Structure, Uncertainty, and the Flexibility of Human Thinking

Human cognition is incredibly flexible, partly because common-sense knowledge is uncertain but highly structured. Probabilistic programming languages (PPLs) provide a formal tool encompassing probabilistic uncertainty and compositional structure. I will show that PPLs allow us to model human reasoning and language understanding. I will describe several experimental studies of reasoning and social cognition. I will then discuss the challenges of universal probabilistic inference, and how deep learning may combine naturally with PPLs to enable learning to do inference.

Speaker: Noah Goodman, Stanford

Room 126