Physics Colloquium: Human learning of semantic networks

The Department of Physics is pleased to welcome Dr. Andrei A. Klishin for an in-person colloquium presentation. Dr. Klishin is a Postdoctoral Researcher with Prof. Dani S. Bassett in the Bioengineering Department at University of Pennsylvania. He works on complex systems, design problems, and human learning using a variety of theoretical and computational tools. He got his Ph.D. from University of Michigan and B.Sc. from MIT, both in Physics.

Abstract

The human mind intuitively decodes sequences of stimuli that arrive from the environment and constructs mental models in the form of complex networks, ranging from language to music to social relationships. But how exactly does one’s brain form a network model of, for example, a linear algebra textbook? I answer this question by weaving together three disciplinary strands. First, neuroscience and psychology elucidate how memories of the sequence are encoded in the brain and what is the pattern of mental errors in the recording. Second, text analysis and network science represent the textbook as a growing weighted network that a human can explore. Third, statistical mechanics and stochastic processes account for the patterns of network exploration. I unify these views by proposing the exposure theory that offers both a parsimonious description of dynamics on a broad class of networks, and a fast and accurate prediction of exploration under the dual pressures of finite learning time and mental errors. While short learning time in general leads to worse learning, textbook structures inherently prioritize the learning of most important concepts and connections. Mental errors in general decrease the precision and recall of learning but can lead to a limited prediction of future edges and a replica symmetry breaking between the learners.

This is an in-person colloquium. If you are interested in attending virtually, please contact phyadmin@syr.edu for the Zoom link.