Embedding Principle: a hierarchical structure of loss landscape of deep neural networks

Abstract

We prove a general Embedding Principle of loss landscape of deep neural networks (NNs) that unravels a hierarchical structure of the loss landscape of NNs, i.e., loss landscape of an NN contains all critical points of all the narrower NNs. This result is obtained by constructing a class of critical embeddings which map any critical point of a narrower NN to a critical point of the target NN with the same output function. By discovering a wide class of general compatible critical embeddings, we provide a gross estimate of the dimension of critical submanifolds embedded from critical points of narrower NNs. We further prove an irreversiblility property of any critical embedding that the number of negative/zero/positive eigenvalues of the Hessian matrix of a critical point may increase but never decrease as an NN becomes wider through the embedding. Using a special realization of general compatible critical embedding, we prove a stringent necessary condition for being a "truly-bad" critical point that never becomes a strict-saddle point through any critical embedding. This result implies the commonplace of strict-saddle points in wide NNs, which may be an important reason underlying the easy optimization of wide NNs widely observed in practice.

References in corpus (10)

• The Loss Surfaces of Multilayer Networks
• A Closer Look at Memorization in Deep Networks
• Stochastic Gradient Descent Optimizes Over-parameterized Deep ReLU Networks
• Towards Understanding Generalization of Deep Learning: Perspective of Loss Landscapes
• The Global Landscape of Neural Networks: An Overview
• Asymmetric Valleys: Beyond Sharp and Flat Local Minima
• Loss Patterns of Neural Networks
• Neural networks are a priori biased towards Boolean functions with low entropy
• Sub-Optimal Local Minima Exist for Neural Networks with Almost All Non-Linear Activations
• Assessing the Bilingual Knowledge Learned by Neural Machine Translation Models