Mapping How AI Learns — Visualizing the Hidden Geometry of Neural Networks
Professor: Gal Mishne
Description: This project explores what actually happens inside a neural network while it learns. Instead of treating an AI model as a “black box,” students take mathematical snapshots of the network’s internal activity during training and map how those internal representations evolve over time. Watching how these points move over time reveals the “learning dynamics” of the network, showing the exact journey the AI takes from random initialization to accurate predictions. Understanding the geometry of learning gives us a diagnostic dashboard for AI. The project will be run concurrently with myself and Alex Cloninger as co-mentors.
Preferred Qualifications: Python programming skills, linear algebra, basic familiarity with machine learning
