Efficient local linearity regularization to overcome catastrophic overfitting

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Rocamora, Elias Abad, Liu, Fanghui, Chrysos, Grigorios G., Olmos, Pablo M., Cevher, Volkan
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911785599107072
author Rocamora, Elias Abad
Liu, Fanghui
Chrysos, Grigorios G.
Olmos, Pablo M.
Cevher, Volkan
author_facet Rocamora, Elias Abad
Liu, Fanghui
Chrysos, Grigorios G.
Olmos, Pablo M.
Cevher, Volkan
contents Catastrophic overfitting (CO) in single-step adversarial training (AT) results in abrupt drops in the adversarial test accuracy (even down to 0%). For models trained with multi-step AT, it has been observed that the loss function behaves locally linearly with respect to the input, this is however lost in single-step AT. To address CO in single-step AT, several methods have been proposed to enforce local linearity of the loss via regularization. However, these regularization terms considerably slow down training due to Double Backpropagation. Instead, in this work, we introduce a regularization term, called ELLE, to mitigate CO effectively and efficiently in classical AT evaluations, as well as some more difficult regimes, e.g., large adversarial perturbations and long training schedules. Our regularization term can be theoretically linked to curvature of the loss function and is computationally cheaper than previous methods by avoiding Double Backpropagation. Our thorough experimental validation demonstrates that our work does not suffer from CO, even in challenging settings where previous works suffer from it. We also notice that adapting our regularization parameter during training (ELLE-A) greatly improves the performance, specially in large $ε$ setups. Our implementation is available in https://github.com/LIONS-EPFL/ELLE .
format Preprint
id arxiv_https___arxiv_org_abs_2401_11618
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Efficient local linearity regularization to overcome catastrophic overfitting
Rocamora, Elias Abad
Liu, Fanghui
Chrysos, Grigorios G.
Olmos, Pablo M.
Cevher, Volkan
Machine Learning
Artificial Intelligence
Cryptography and Security
Catastrophic overfitting (CO) in single-step adversarial training (AT) results in abrupt drops in the adversarial test accuracy (even down to 0%). For models trained with multi-step AT, it has been observed that the loss function behaves locally linearly with respect to the input, this is however lost in single-step AT. To address CO in single-step AT, several methods have been proposed to enforce local linearity of the loss via regularization. However, these regularization terms considerably slow down training due to Double Backpropagation. Instead, in this work, we introduce a regularization term, called ELLE, to mitigate CO effectively and efficiently in classical AT evaluations, as well as some more difficult regimes, e.g., large adversarial perturbations and long training schedules. Our regularization term can be theoretically linked to curvature of the loss function and is computationally cheaper than previous methods by avoiding Double Backpropagation. Our thorough experimental validation demonstrates that our work does not suffer from CO, even in challenging settings where previous works suffer from it. We also notice that adapting our regularization parameter during training (ELLE-A) greatly improves the performance, specially in large $ε$ setups. Our implementation is available in https://github.com/LIONS-EPFL/ELLE .
title Efficient local linearity regularization to overcome catastrophic overfitting
topic Machine Learning
Artificial Intelligence
Cryptography and Security
url https://arxiv.org/abs/2401.11618