Low Dimensional Landscape Hypothesis is True: DNNs can be Trained in Tiny Subspaces
Primary LanguagePythonMIT LicenseMIT