Scam Alert

Scam Alert

Please verify and be careful about any phishing and scam attempts from external companies.
All conferences and research programs at IML are free of charge.
We will not ask you for any payments regarding your accommodation or travel arrangements

Deoliveira: Weak-to-Strong generalization for random features networks using random matrix theory

Date: 2026-07-15

Time: 11:20 - 11:55

Zoom link: https://kva-se.zoom.us/j/9217561890

Speaker
Deoliveira, McGill University

Abstract
In this work, under suitable assumptions, we prove the occurrence of the Weak-to-Strong generalization phenomenon, where a strong model learns from labels generated by a weak model and still outperforms it. We study this phenomenon in a two-layer random features model, where the model strength is determined by the number of neurons. Using tools from random matrix theory, in particular deterministic equivalents obtained through a self-consistent Dyson equation, we characterize the population errors of the weak teacher and the strong student. This allows us to find the exact exponent characterizing how the strong model’s population error improves relative to the weak model’s error for linear and ReLU activation functions and polynomial target functions. This is a joint work with Elliot Paquette.