Feedback Gradient Descent: Efficient and Stable Optimization with Orthogonality for DNNs

Abstract

The optimization with orthogonality has been shown useful in training deep neural networks (DNNs). To impose orthogonality on DNNs, both computational efficiency and stability are important. However, existing methods utilizing Riemannian optimization or hard constraints can only ensure stability while those using soft constraints can only improve efficiency. In this paper, we propose a novel method, named Feedback Gradient Descent (FGD), to our knowledge, the first work showing high efficiency and stability simultaneously. FGD induces orthogonality based on the simple yet indispensable Euler discretization of a continuous-time dynamical system on the tangent bundle of the Stiefel manifold. In particular, inspired by a numerical integration method on manifolds called Feedback Integrators, we propose to instantiate it on the tangent bundle of the Stiefel manifold for the first time. In the extensive image classification experiments, FGD comprehensively outperforms the existing state-of-the-art methods in terms of accuracy, efficiency, and stability.

References in corpus (16)

• Very Deep Convolutional Networks for Large-Scale Image Recognition
• Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
• PyTorch: An Imperative Style, High-Performance Deep Learning Library
• On the difficulty of training Recurrent Neural Networks
• A Simple Way to Initialize Recurrent Networks of Rectified Linear Units
• Neural Photo Editing with Introspective Adversarial Networks
• Full-Capacity Unitary Recurrent Neural Networks
• Trivializations for Gradient-Based Optimization on Manifolds
• Efficient Riemannian Optimization on the Stiefel Manifold via the Cayley Transform
• Deep Isometric Learning for Visual Recognition
• Projection Based Weight Normalization for Deep Neural Networks
• Orthogonalizing Convolutional Layers with the Cayley Transform
• Convolutional Normalization: Improving Deep Convolutional Network Robustness and Training
• Adaptive and Momentum Methods on Manifolds Through Trivializations
• Eigenvalue Normalized Recurrent Neural Networks for Short Term Memory
• Tracking Controller Design for Satellite Attitude Under Unknown Constant Disturbance Using Stable Embedding