Stop Wasting My Gradients: Practical SVRG

Abstract

We present and analyze several strategies for improving the performance of stochastic variance-reduced gradient (SVRG) methods. We first show that the convergence rate of these methods can be preserved under a decreasing sequence of errors in the control variate, and use this to derive variants of SVRG that use growing-batch strategies to reduce the number of gradient calculations required in the early iterations. We further (i) show how to exploit support vectors to reduce the number of gradient computations in the later iterations, (ii) prove that the commonly-used regularized SVRG iteration is justified and improves the convergence rate, (iii) consider alternate mini-batch selection strategies, and (iv) consider the generalization error of the method.

References in corpus (1)

• Piecewise linear regularized solution paths

Cited by in corpus (40)

• Federated Optimization: Distributed Machine Learning for On-Device Intelligence
• Online Batch Selection for Faster Training of Neural Networks
• Barzilai-Borwein Step Size for Stochastic Gradient Descent
• Non-convex Finite-Sum Optimization Via SCSG Methods
• Why Random Reshuffling Beats Stochastic Gradient Descent
• AdaBatch: Adaptive Batch Sizes for Training Deep Neural Networks
• Convergence guarantees for RMSProp and ADAM in non-convex optimization and an empirical comparison to Nesterov acceleration
• ProxSARAH: An Efficient Algorithmic Framework for Stochastic Composite Nonconvex Optimization
• Stochastic L-BFGS: Improved Convergence Rates and Practical Acceleration Strategies
• Sample Efficient Policy Gradient Methods with Recursive Variance Reduction
• Stochastic Polyak Step-size for SGD: An Adaptive Learning Rate for Fast Convergence
• Coupling Adaptive Batch Sizes with Learning Rates
• Stochastic Variance Reduction Methods for Saddle-Point Problems
• Less than a Single Pass: Stochastically Controlled Stochastic Gradient Method
• An Improved Convergence Analysis of Stochastic Variance-Reduced Policy Gradient
• Starting Small -- Learning with Adaptive Sample Sizes
• Reducing the variance in online optimization by transporting past gradients
• Towards closing the gap between the theory and practice of SVRG
• History-Gradient Aided Batch Size Adaptation for Variance Reduced Algorithms
• FasTrCaps: An Integrated Framework for Fast yet Accurate Training of Capsule Networks
• Trading-off variance and complexity in stochastic gradient descent
• A Convergence Analysis for A Class of Practical Variance-Reduction Stochastic Gradient MCMC
• An Analysis of Stochastic Variance Reduced Gradient for Linear Inverse Problems
• Fast and Furious Convergence: Stochastic Second Order Methods under Interpolation
• Variance reduction for Riemannian non-convex optimization with batch size adaptation
• SCOPE: Scalable Composite Optimization for Learning on Spark
• A Generalizable Approach to Learning Optimizers
• Approximate Newton-based statistical inference using only stochastic gradients
• A Fast, Principled Working Set Algorithm for Exploiting Piecewise Linear Structure in Convex Problems
• Accelerating Convergence of Replica Exchange Stochastic Gradient MCMC via Variance Reduction
• SVRG for Policy Evaluation with Fewer Gradient Evaluations
• Reducing Runtime by Recycling Samples
• SONIA: A Symmetric Blockwise Truncated Optimization Algorithm
• A Stochastic Large-scale Machine Learning Algorithm for Distributed Features and Observations
• SGD with Variance Reduction beyond Empirical Risk Minimization
• The Lingering of Gradients: Theory and Applications
• A Multilevel Approach to Training
• Asynchronous Parallel Stochastic Quasi-Newton Methods
• Some Limit Properties of Markov Chains Induced by Stochastic Recursive Algorithms
• Continuous-time Models for Stochastic Optimization Algorithms