Daniel Wurgaft | A Resource Rational Analysis of Representations: Geometry, Inference, Capacity

Abstract

Neural systems process information by forming representations of the world. Studies of representational geometry have revealed remarkably structured representations in biological and artificial systems, as well as surprising alignment in representations across systems. Why do representations across systems converge to similar structures, how do these structures encode concepts and uncertainty over them, and how are they shaped by resource constraints?

Here, we tackle these questions in neural networks through the lens of resource rationality, asking what representations are optimal for solving a task, given dimensionality and linear readout constraints. We derive a notion of optimal linear representations, and show that these emerge across varied architectures and tasks, providing a simple explanation for representational convergence as alignment toward the optimal solution. This shared treatment motivates a normative formalization of conceptual spaces that generalizes the linear representation hypothesis. We then extend this account to uncertainty and inference over conceptual spaces, asking how representations should change in response to in-context evidence, and finally to resource-constrained settings involving representational interference.

About the Center for Neural Data Science Seminar Series 

The Center for Neural Data Science Seminar Series is a platform for trainees across campus to share  insights and innovative approaches that bridge the gap between neuroscience and data science. 

As neuroscience continues to generate vast amounts of data—from intricate neural circuit maps to large-scale brain activity recordings—the need for interdisciplinary expertise in data science, statistics, and engineering has never been more critical. 

The Center for Neural Data Science mission is to advance brain research through the development of cutting-edge analytical methodologies and collaborative approaches. Stanford's affiliates are invited to join this vibrant community dedicated to transformative discoveries.

This seminar series is only offered in person. 

Happy Hour/Reception for Attendees: 

Seminars attendees are invited to stay for a happy hour reception after the seminar to continue the conversation with the speaker.

Sign up to learn about all our upcoming events