OntheSymmetriesofDeepLearningModelsand theirInternalRepresentations