Creating New Perceptual Descriptions During Category Learning

This lecture by Robert Goldstone from Indiana University Bloomington explores the challenges of Bongard problems in category learning, focusing on distinguishing visual scenes that fall into different categories. Learn about Physical Bongard Problems (PBPs), which require understanding the physical spatial dynamics implicit in depicted scenes. Discover the PATHS (Perceiving And Testing Hypotheses on Structures) computational model that solves many PBPs, and see how it compares to human performance on the same problems. Explore the core theoretical commitments of PATHS that exemplify human open-ended category learning, including continual perception of new scene descriptions, context-dependent perceptual processes, hypothesis construction through logical expressions, and bi-directional interactions between perception and hypothesis construction. This 44-minute talk from the Santa Fe Institute provides valuable insights into how both humans and machines develop new perceptual descriptions during the category learning process.