AI Dynamics

Global AI News Aggregator

Training Degeneracy in Rectangular Weight Matrices Without Hadamard

This is the training degeneracy that @jiawzhao is talking about if you only do identity initialization and not Hadamard for rectangular weight matrices

→ View original post on X — @animaanandkumar,

Commentaires

Leave a Reply

Your email address will not be published. Required fields are marked *