Poster
in
Workshop: Symmetry and Geometry in Neural Representations

The Representations of Deep Neural Networks Trained on Dihedral Group Multiplication

Gavin McCracken · Sihui Wei · Gabriela Moisescu-Pareja · Harley Wiltzer · Jonathan Love

Project Page [ OpenReview]

Abstract

We find coset and approximate coset circuits play a key role in how multilayer perceptrons learn dihedral group multiplication, consistent with recent findings on modular addition.

Chat is not available.