Cognitive Science Colloquium: Keith Holyoak

Description

Dr. Keith Holyoak, Distinguished Professor in the Department of Psychology at UCLA, will present on "Semantic Relations in Minds, Brains, and Machines." Children and adults are able to learn abstract semantic relations, such as synonymy and antonymy, which cannot be defined in terms of specific features of individual objects. Such relations are central to human intelligence, underlying the distinctively human ability to reason by analogy across situations that initially appear highly dissimilar.

New algorithms in machine learning can automatically generate rich semantic vectors (word embeddings) representing the meanings of individual words. We are developing a computer model (Bayesian Analogy with Relational Transformations) that can extract deeper representations of abstract semantic relations from such non-relational inputs.

Analogical reasoning emerges as a natural consequence of relation learning. BART's representations predict patterns of similarity among neural signals triggered by a variety of semantic relations. We are extending the approach to visual inputs. Core properties of high-level intelligence may emerge from operations that transform featural representations into meaningful relations.