Stochastic Gradient Descent with Biased but Consistent Gradient Estimators

Abstract

Stochastic gradient descent (SGD), which dates back to the 1950s, is one of the most popular and effective approaches for performing stochastic optimization. Research on SGD resurged recently in machine learning for optimizing convex loss functions and training nonconvex deep neural networks. The theory assumes that one can easily compute an unbiased gradient estimator, which is usually the case due to the sample average nature of empirical risk minimization. There exist, however, many scenarios (e.g., graphs) where an unbiased estimator may be as expensive to compute as the full gradient because training examples are interconnected. Recently, Chen et al. (2018) proposed using a consistent gradient estimator as an economic alternative. Encouraged by empirical success, we show, in a general setting, that consistent estimators result in the same convergence behavior as do unbiased ones. Our analysis covers strongly convex, convex, and nonconvex objectives. We verify the results with illustrative experiments on synthetic and real-world data. This work opens several new research directions, including the development of more efficient SGD updates with consistent estimators and the design of efficient training algorithms for large-scale graphs.

Companion codes are at https://github.com/jiechenjiechen/FastGCN-matlab

References in corpus (2)

• Inductive Representation Learning on Large Graphs
• Neural Message Passing for Quantum Chemistry

Cited by in corpus (12)

• Training Gaussian Boson Sampling Distributions
• Adaptively Truncating Backpropagation Through Time to Control Gradient Bias
• Accelerating Training and Inference of Graph Neural Networks with Fast Sampling and Pipelining
• On the Convergence of SGD with Biased Gradients
• Extreme Compressed Sensing of Poisson Rates from Multiple Measurements
• Differentiable Visual Computing
• Gaussian Process Inference Using Mini-batch Stochastic Gradient Descent: Convergence Guarantees and Empirical Benefits
• The Sharpe predictor for fairness in machine learning
• On the Importance of Sampling in Training GCNs: Tighter Analysis and Variance Reduction
• A maximum-entropy approach to off-policy evaluation in average-reward MDPs
• Sample Complexity of Estimating the Policy Gradient for Nearly Deterministic Dynamical Systems
• A Variance Controlled Stochastic Method with Biased Estimation for Faster Non-convex Optimization