85-726 Reverse Engineering the Mind: Perception in Brains and Machines

12 units, Carnegie Mellon University

How can we determine if the internal structure of an artificial neural network truly captures the principles of biological computation? This course takes a forensic approach to scrutinizing the representations developed by both brains and machines. We will explore the structural alignment of these systems using the tools of representational geometry—such as RSA and manifold analysis—to map how sensory information is transformed into neural code. A central focus will be the use of computational probes to expose the latent constraints of perception. We will analyze perceptual illusions as "hacks" of the biological brain and compare them to model metamers and adversarial perturbations that "hack" artificial systems. By examining where these models align with or spectacularly fail human psychophysics, students will gain a deep understanding of the requirements for building truly brain-like AI. This is a methods-forward seminar for students interested in the rigorous evaluation of NeuroAI.

Prerequisites: none listed

Offered: Fall 2026

Courseweave needs JavaScript to show sections, workloads and planning tools.