Beyond Categorical Label Representations for Image Classification

Boyuan Chen, Yu Li, Sunand Raghupathi +1

We find that the way we choose to represent data labels can have a profound effect on the quality of trained models. For example, training an image classifier to regress audio labe…