#diagnosis

Question 1: Select all that apply

Three training runs end with these losses:

Which conclusions are justified? Select all that apply.

Run B's large gap between training and validation loss points to overfitting1
Run A's high training loss could come from too little capacity or from a poorly working optimization (e.g. an unsuitable learning rate)2
Run A must be too small, so adding layers is the only remedy3
The regularization strength of run C should be tuned by comparing test-set losses4