Carathéodory Sampling for Stochastic Gradient Descent

Abstract

Many problems require to optimize empirical risk functions over large data sets. Gradient descent methods that calculate the full gradient in every descent step do not scale to such datasets. Various flavours of Stochastic Gradient Descent (SGD) replace the expensive summation that computes the full gradient by approximating it with a small sum over a randomly selected subsample of the data set that in turn suffers from a high variance. We present a different approach that is inspired by classical results of Tchakaloff and Carathéodory about measure reduction. These results allow to replace an empirical measure with another, carefully constructed probability measure that has a much smaller support, but can preserve certain statistics such as the expected gradient. To turn this into scalable algorithms we firstly, adaptively select the descent steps where the measure reduction is carried out; secondly, we combine this with Block Coordinate Descent so that measure reduction can be done very cheaply. This makes the resulting methods scalable to high-dimensional spaces. Finally, we provide an experimental validation and comparison.

References in corpus (9)

• Adam: A Method for Stochastic Optimization
• An overview of gradient descent optimization algorithms
• A Stochastic Gradient Method with an Exponential Convergence Rate for Finite Training Sets
• Parallel Coordinate Descent for L1-Regularized Loss Minimization
• A Fast Active Set Block Coordinate Descent Algorithm for ${\ell }_1$-regularized least squares
• Global Convergence of Arbitrary-Block Gradient Methods for Generalized Polyak-Łojasiewicz Functions
• Greedy Block Coordinate Descent for Large Scale Gaussian Process Regression
• Fast and Accurate Least-Mean-Squares Solvers
• Greedy Block Coordinate Descent (GBCD) Method for High Dimensional Quadratic Programs

Cited by in corpus (1)

• A Randomized Algorithm to Reduce the Support of Discrete Measures