Saved in:
Bibliographic Details
Main Authors: Malviya, Pranshu, Mordido, Gonçalo, Baratin, Aristide, Harikandeh, Reza Babanezhad, Huang, Jerry, Lacoste-Julien, Simon, Pascanu, Razvan, Chandar, Sarath
Format: Preprint
Published: 2023
Subjects:
Online Access:https://arxiv.org/abs/2307.09638
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911921219829760
author Malviya, Pranshu
Mordido, Gonçalo
Baratin, Aristide
Harikandeh, Reza Babanezhad
Huang, Jerry
Lacoste-Julien, Simon
Pascanu, Razvan
Chandar, Sarath
author_facet Malviya, Pranshu
Mordido, Gonçalo
Baratin, Aristide
Harikandeh, Reza Babanezhad
Huang, Jerry
Lacoste-Julien, Simon
Pascanu, Razvan
Chandar, Sarath
contents Adaptive gradient-based optimizers, notably Adam, have left their mark in training large-scale deep learning models, offering fast convergence and robustness to hyperparameter settings. However, they often struggle with generalization, attributed to their tendency to converge to sharp minima in the loss landscape. To address this, we propose a new memory-augmented version of Adam that encourages exploration towards flatter minima by incorporating a buffer of critical momentum terms during training. This buffer prompts the optimizer to overshoot beyond narrow minima, promoting exploration. Through comprehensive analysis in simple settings, we illustrate the efficacy of our approach in increasing exploration and bias towards flatter minima. We empirically demonstrate that it can improve model performance for image classification on ImageNet and CIFAR10/100, language modelling on Penn Treebank, and online learning tasks on TinyImageNet and 5-dataset. Our code is available at \url{https://github.com/chandar-lab/CMOptimizer}.
format Preprint
id arxiv_https___arxiv_org_abs_2307_09638
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Promoting Exploration in Memory-Augmented Adam using Critical Momenta
Malviya, Pranshu
Mordido, Gonçalo
Baratin, Aristide
Harikandeh, Reza Babanezhad
Huang, Jerry
Lacoste-Julien, Simon
Pascanu, Razvan
Chandar, Sarath
Machine Learning
Artificial Intelligence
Adaptive gradient-based optimizers, notably Adam, have left their mark in training large-scale deep learning models, offering fast convergence and robustness to hyperparameter settings. However, they often struggle with generalization, attributed to their tendency to converge to sharp minima in the loss landscape. To address this, we propose a new memory-augmented version of Adam that encourages exploration towards flatter minima by incorporating a buffer of critical momentum terms during training. This buffer prompts the optimizer to overshoot beyond narrow minima, promoting exploration. Through comprehensive analysis in simple settings, we illustrate the efficacy of our approach in increasing exploration and bias towards flatter minima. We empirically demonstrate that it can improve model performance for image classification on ImageNet and CIFAR10/100, language modelling on Penn Treebank, and online learning tasks on TinyImageNet and 5-dataset. Our code is available at \url{https://github.com/chandar-lab/CMOptimizer}.
title Promoting Exploration in Memory-Augmented Adam using Critical Momenta
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2307.09638