No Wrong Turns: The Simple Geometry Of Neural Networks Optimization Paths

arXiv2306.11922MaRDI QIDQ6440926

Author name not available (Why is that?)

Publication date: 20 June 2023

Abstract: Understanding the optimization dynamics of neural networks is necessary for closing the gap between theory and practice. Stochastic first-order optimization algorithms are known to efficiently locate favorable minima in deep neural networks. This efficiency, however, contrasts with the non-convex and seemingly complex structure of neural loss landscapes. In this study, we delve into the fundamental geometric properties of sampled gradients along optimization paths. We focus on two key quantities, which appear in the restricted secant inequality and error bound. Both hold high significance for first-order optimization. Our analysis reveals that these quantities exhibit predictable, consistent behavior throughout training, despite the stochasticity induced by sampling minibatches. Our findings suggest that not only do optimization trajectories never encounter significant obstacles, but they also maintain stable dynamics during the majority of training. These observed properties are sufficiently expressive to theoretically guarantee linear convergence and prescribe learning rate schedules mirroring empirical practices. We conduct our experiments on image classification, semantic segmentation and language modeling across different batch sizes, network architectures, datasets, optimizers, and initialization seeds. We discuss the impact of each factor. Our work provides novel insights into the properties of neural network loss functions, and opens the door to theoretical frameworks more relevant to prevalent practice.

Has companion code repository: https://github.com/hiroki11x/losslandscapegeometry

This page was built for publication: No Wrong Turns: The Simple Geometry Of Neural Networks Optimization Paths

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6440926)