On the training dynamics of deep networks with regularization
arXiv:2006.08643
Abstract
We study the role of regularization in deep learning, and uncover simple relations between the performance of the model, the coefficient, the learning rate, and the number of training steps. These empirical relations hold when the network is overparameterized. They can be used to predict the optimal regularization parameter of a given model. In addition, based on these observations we propose a dynamical schedule for the regularization parameter that improves performance and speeds up training. We test these proposals in modern image classification settings. Finally, we show that these empirical relations can be understood theoretically in the context of infinitely wide networks. We derive the gradient flow dynamics of such networks, and compare the role of regularization in this context with that of linear models.
arXiv comment10+12 pages, 5+10 figures. Updated to match NeurIPS version
References in corpus (15)
- Neural Tangent Kernel: Convergence and Generalization in Neural Networks
- Understanding deep learning requires rethinking generalization
- On Exact Computation with an Infinitely Wide Neural Net
- Exploring Generalization in Deep Learning
- L2 Regularization versus Batch and Weight Normalization
- Scaling description of generalization with number of parameters in deep learning
- Bayesian Deep Convolutional Networks with Many Channels are Gaussian Processes
- Empirical Analysis of the Hessian of Over-Parametrized Neural Networks
- Gradient Descent Happens in a Tiny Subspace
- Norm matters: efficient and accurate normalization schemes in deep networks
- The large learning rate phase of deep learning: the catapult mechanism
- Three Mechanisms of Weight Decay Regularization
- An Exponential Learning Rate Schedule for Deep Learning
- Reconciling Modern Deep Learning with Traditional Optimization Analyses: The Intrinsic Learning Rate
- On the asymptotics of wide networks with polynomial activations