On the fast convergence of minibatch heavy ball momentum
arXiv:2206.07553 · doi:10.1093/imanum/drae033
Abstract
Simple stochastic momentum methods are widely used in machine learning optimization, but their good practical performance is at odds with an absence of theoretical guarantees of acceleration in the literature. In this work, we aim to close the gap between theory and practice by showing that stochastic heavy ball momentum retains the fast linear rate of (deterministic) heavy ball momentum on quadratic optimization problems, at least when minibatching with a sufficiently large batch size. The algorithm we study can be interpreted as an accelerated randomized Kaczmarz algorithm with minibatching and heavy ball momentum. The analysis relies on carefully decomposing the momentum transition matrix, and using new spectral norm concentration bounds for products of independent random matrices. We provide numerical illustrations demonstrating that our bounds are reasonably sharp.
arXiv commentupdate to match journal version
References in corpus (7)
- Paved with Good Intentions: Analysis of a Randomized Block Kaczmarz Method
- Randomized Kaczmarz solver for noisy linear systems
- Adaptive Bound Optimization for Online Convex Optimization
- A Simple Practical Accelerated Method for Finite Sums
- A Simple Convergence Proof of Adam and Adagrad
- Linearly convergent stochastic heavy ball method for minimizing generalization error
- Trajectory of Mini-Batch Momentum: Batch Size Saturation and Convergence in High Dimensions