■ Title : Creating Something Different: Similarity, Contrast, and Representativeness
■ Speaker : Prof. Joseph Austerweil @ WISC (https://alab.psych.wisc.edu/ )
■ Time : 2023 March 9th 11:00 ~ 12:00
■ Location: Hybrid
- Online: https://xinics.zoom.us/j/84026590333 (Passcode: 53069072)
- In-person : 제2공학관 26223호
■ Language: English speech & English slides
Recent advances in machine learning have demonstrated their power in generating new content. One pressing question is whether they generate in a human-like manner. The ability to generate new concepts and ideas is among the most fascinating aspects of human cognition, but we do not have a strong understanding of the cognitive processes and representations underlying concept generation. In this talk, I will discuss the generation of new categories using the computational and behavioral toolkit of traditional artificial category learning. Previous work in this domain has focused on how the statistical structure of known categories generalizes to generated categories, overlooking whether (and if so, how) contrast between the known and generated categories is a factor. I will present experiments demonstrating that contrast between what is known and what is created informs human generation. I will present two novel approaches to modeling category contrast: one focused on dissimilarity of a kernel density estimate and another on the Bayesian representativeness heuristic. I will conclude with a discussion of how integrating this work with state-of-the-art generative machine learning could yield insights in both human and machine intelligence.
Joe Austerweil (firstname.lastname@example.org) is an associate professor in the Department of Psychology and Computer Science (affiliate) at the University of Wisconsin-Madison. He uses computational models and behavioral experiments to understand how people reason and make decisions.