Publication

Extrapolation for Large-batch Training in Deep Learning

Martin Jaggi, Sebastian Urban Stich, Tao Lin, Lingjing Kong
2020
Conference Papers

Résumé

Deep learning networks are typically trained by Stochastic Gradient Descent (SGD) methods that iteratively improve the model parameters by estimating a gradient on a very small fraction of the training data. A major roadblock faced when increasing the batch size to a substantial fraction of the training data for reducing training time is the persistent degradation in performance (generalization gap). To address this issue, recent work propose to add small perturbations to the model parameters when computing the stochastic gradients and report improved generalization performance due to smoothing effects. However, this approach is poorly understood; it requires often model-specific noise and fine-tuning. To alleviate these drawbacks, we propose to use instead computationally efficient extrapolation (extragradient) to stabilize the optimization trajectory while still benefiting from smoothing to avoid sharp minima. This principled approach is well grounded from an optimization perspective and we show that a host of variations can be covered in a unified framework that we propose. We prove the convergence of this novel scheme and rigorously evaluate its empirical performance on ResNet, LSTM, and Transformer. We demonstrate that in a variety of experiments the scheme allows scaling to much larger batch sizes than before whilst reaching or surpassing SOTA accuracy.

Source officielle

https://infoscience.epfl.ch/entities/publication/b855d4dd-bede-4a56-a0c7-37189fe3bd04

À propos de ce résultat

Cette page est générée automatiquement et peut contenir des informations qui ne sont pas correctes, complètes, à jour ou pertinentes par rapport à votre recherche. Il en va de même pour toutes les autres pages de ce site. Veillez à vérifier les informations auprès des sources officielles de l'EPFL.

Martin Jaggi, Sebastian Urban Stich, Tao Lin, Lingjing Kong
2020
Conference Papers

Résumé

Source officielle

https://infoscience.epfl.ch/entities/publication/b855d4dd-bede-4a56-a0c7-37189fe3bd04

À propos de ce résultat

Concepts associés (32)

Apprentissage profond

L'apprentissage profond ou apprentissage en profondeur (en anglais : deep learning, deep structured learning, hierarchical learning) est un sous-domaine de l’intelligence artificielle qui utilise des réseaux neuronaux pour résoudre des tâches complexes grâce à des architectures articulées de différentes transformations non linéaires. Ces techniques ont permis des progrès importants et rapides dans les domaines de l'analyse du signal sonore ou visuel et notamment de la reconnaissance faciale, de la reconnaissance vocale, de la vision par ordinateur, du traitement automatisé du langage.

Algorithme du gradient stochastique

L'algorithme du gradient stochastique est une méthode de descente de gradient (itérative) utilisée pour la minimisation d'une fonction objectif qui est écrite comme une somme de fonctions différentiables. À la fois l'estimation statistique et l'apprentissage automatique s'intéressent au problème de la minimisation d'une fonction objectif qui a la forme d'une somme : où le paramètre qui minimise doit être estimé. Chacune des fonctions est généralement associée avec la -ème observation de l'ensemble des données (utilisées pour l'apprentissage).

Batch normalization

Batch normalization (also known as batch norm) is a method used to make training of artificial neural networks faster and more stable through normalization of the layers' inputs by re-centering and re-scaling. It was proposed by Sergey Ioffe and Christian Szegedy in 2015. While the effect of batch normalization is evident, the reasons behind its effectiveness remain under discussion. It was believed that it can mitigate the problem of internal covariate shift, where parameter initialization and changes in the distribution of the inputs of each layer affect the learning rate of the network.

Afficher plus

Publications associées (32)

Understanding generalization and robustness in modern deep learning

Maksym Andriushchenko

In this thesis, we study two closely related directions: robustness and generalization in modern deep learning. Deep learning models based on empirical risk minimization are known to be often non-robust to small, worst-case perturbations known as adversari ...

EPFL2024

On the Generalization of Stochastic Gradient Descent with Momentum

Volkan Cevher, Kimon Antonakopoulos

While momentum-based accelerated variants of stochastic gradient descent (SGD) are widely used when training machine learning models, there is little theoretical understanding on the generalization error of such methods. In this work, we first show that th ...

Microtome Publishing2024

Optimization Algorithms for Decentralized, Distributed and Collaborative Machine Learning

Anastasiia Koloskova

Distributed learning is the key for enabling training of modern large-scale machine learning models, through parallelising the learning process. Collaborative learning is essential for learning from privacy-sensitive data that is distributed across various ...

EPFL2024

Afficher plus