The Implicit Bias of Gradient Descent on Separable Multiclass Data

However, there is a noticeable gap in analysis for multiclass classification, with only a handful of results which themselves are restricted to the cross-entropy loss.