Orthogonal Over-Parameterized Training
arXiv:2004.04690
Abstract
The inductive bias of a neural network is largely determined by the architecture and the training algorithm. To achieve good generalization, how to effectively train a neural network is of great importance. We propose a novel orthogonal over-parameterized training (OPT) framework that can provably minimize the hyperspherical energy which characterizes the diversity of neurons on a hypersphere. By maintaining the minimum hyperspherical energy during training, OPT can greatly improve the empirical generalization. Specifically, OPT fixes the randomly initialized weights of the neurons and learns an orthogonal transformation that applies to these neurons. We consider multiple ways to learn such an orthogonal transformation, including unrolling orthogonalization algorithms, applying orthogonal parameterization, and designing orthogonality-preserving gradient descent. For better scalability, we propose the stochastic OPT which performs orthogonal transformation stochastically for partial dimensions of neurons. Interestingly, OPT reveals that learning a proper coordinate system for neurons is crucial to generalization. We provide some insights on why OPT yields better generalization. Extensive experiments validate the superiority of OPT over the standard training.
arXiv commentCVPR 2021 Oral (43 Pages, Substantial Update from v3, Typos Fixed from v5)
References in corpus (22)
- Adam: A Method for Stochastic Optimization
- Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
- Semi-Supervised Classification with Graph Convolutional Networks
- Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks
- ADADELTA: An Adaptive Learning Rate Method
- Prototypical Networks for Few-shot Learning
- Learning Face Representation from Scratch
- On the Convergence of Adam and Beyond
- A Convergence Theory for Deep Learning via Over-Parameterization
- Improving Generalization Performance by Switching from Adam to SGD
- A Closer Look at Few-shot Classification
- No bad local minima: Data independent training error guarantees for multilayer neural networks
- Full-Capacity Unitary Recurrent Neural Networks
- Orthogonal Weight Normalization: Solution to Optimization over Multiple Dependent Stiefel Manifolds in Deep Neural Networks
- Tunable Efficient Unitary Neural Networks (EUNN) and their application to RNNs
- Cheap Orthogonal Constraints in Neural Networks: A Simple Parametrization of the Orthogonal and Unitary Group
- Diverse Neural Network Learns True Target Functions
- Recurrent Orthogonal Networks and Long-Memory Tasks
- Efficient Riemannian Optimization on the Stiefel Manifold via the Cayley Transform
- Deep Isometric Learning for Visual Recognition
- Neural Similarity Learning
- One Bit Sensing, Discrepancy, and Stolarsky Principle