Saved in:
Bibliographic Details
Main Authors: Salaün, Corentin, Huang, Xingchang, Georgiev, Iliyan, Mitra, Niloy J., Singh, Gurprit
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2407.15525
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909467949400064
author Salaün, Corentin
Huang, Xingchang
Georgiev, Iliyan
Mitra, Niloy J.
Singh, Gurprit
author_facet Salaün, Corentin
Huang, Xingchang
Georgiev, Iliyan
Mitra, Niloy J.
Singh, Gurprit
contents We introduce a theoretical and practical framework for efficient importance sampling of mini-batch samples for gradient estimation from single and multiple probability distributions. To handle noisy gradients, our framework dynamically evolves the importance distribution during training by utilizing a self-adaptive metric. Our framework combines multiple, diverse sampling distributions, each tailored to specific parameter gradients. This approach facilitates the importance sampling of vector-valued gradient estimation. Rather than naively combining multiple distributions, our framework involves optimally weighting data contribution across multiple distributions. This adapted combination of multiple importance yields superior gradient estimates, leading to faster training convergence. We demonstrate the effectiveness of our approach through empirical evaluations across a range of optimization tasks like classification and regression on both image and point cloud datasets.
format Preprint
id arxiv_https___arxiv_org_abs_2407_15525
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Multiple Importance Sampling for Stochastic Gradient Estimation
Salaün, Corentin
Huang, Xingchang
Georgiev, Iliyan
Mitra, Niloy J.
Singh, Gurprit
Machine Learning
We introduce a theoretical and practical framework for efficient importance sampling of mini-batch samples for gradient estimation from single and multiple probability distributions. To handle noisy gradients, our framework dynamically evolves the importance distribution during training by utilizing a self-adaptive metric. Our framework combines multiple, diverse sampling distributions, each tailored to specific parameter gradients. This approach facilitates the importance sampling of vector-valued gradient estimation. Rather than naively combining multiple distributions, our framework involves optimally weighting data contribution across multiple distributions. This adapted combination of multiple importance yields superior gradient estimates, leading to faster training convergence. We demonstrate the effectiveness of our approach through empirical evaluations across a range of optimization tasks like classification and regression on both image and point cloud datasets.
title Multiple Importance Sampling for Stochastic Gradient Estimation
topic Machine Learning
url https://arxiv.org/abs/2407.15525