A computational framework for complex latent structures

Modern societies are highly complex. Successfully navigating them requires understanding how people relate to one another. For example, who occupies what role? Who is influential in the group? Who is the best person to ask for help or advice? Who is friends with whom? The goal of this project is to understand the psychological processes that allow humans to learn these kinds of relationships through observing patterns of interactions between people.

We refer to the process of inferring relationships between people as “structure learning.” This project investigates structure learning via experiments in which people observe patterns of interaction between several people and are asked to report the kinds of relationships that exist between the people, across many different types of contexts and places. These experiments are also linked to a computational model that makes precise predictions about human responses, allowing us to test formal theories of the psychological processes involved.

Aim 1 develops a theory of how humans infer and reason about underlying group behavior, implemented as a computational model that makes precise and testable predictions. Aim 2 conducts a series of experiments that investigate human judgments and uses them to understand and refine the computational account. This interdisciplinary project integrates work from machine learning, cognitive science, and social psychology, and supports training and research of postdocs, graduate students, and advanced undergraduates across disciplines.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.