Training Deep Networks without Learning Rates Through Coin Betting
arXiv:1705.07795
Abstract
Deep learning methods achieve state-of-the-art performance in many application scenarios. Yet, these methods require a significant amount of hyperparameters tuning in order to achieve the best results. In particular, tuning the learning rates in the stochastic optimization process is still one of the main bottlenecks. In this paper, we propose a new stochastic gradient descent procedure for deep networks that does not require any learning rate setting. Contrary to previous methods, we do not adapt the learning rates nor we make use of the assumed curvature of the objective function. Instead, we reduce the optimization process to a game of betting on a coin and propose a learning-rate-free optimal algorithm for this scenario. Theoretical convergence is proven for convex and quasi-convex functions and empirical evidence shows the advantage of our algorithm over popular stochastic gradient algorithms.
arXiv commentCamera-ready version for NIPS 2017
References in corpus (13)
- ADADELTA: An Adaptive Learning Rate Method
- Caffe: Convolutional Architecture for Fast Feature Embedding
- Recurrent Neural Network Regularization
- No More Pesky Learning Rates
- Practical recommendations for gradient-based training of deep architectures
- Gradient Descent Learns Linear Dynamical Systems
- Unconstrained Online Linear Learning in Hilbert Spaces: Minimax Algorithms and Normal Approximations
- No-Regret Algorithms for Unconstrained Online Convex Optimization
- Less Regret via Online Conditioning
- Coin Betting and Parameter-Free Online Learning
- Online Convex Optimization with Unconstrained Domains and Losses
- Online Learning Without Prior Information
- Scale-Free Algorithms for Online Linear Optimization