Under mild assumptions, we investigate the structure of loss landscape o...
We propose an optimistic estimate to evaluate the best possible fitting
...
Models with nonlinear architectures/parameterizations such as deep neura...
In recent years, understanding the implicit regularization of neural net...
share