ImpNet: Imperceptible and blackbox-undetectable backdoors in compiled neural networks

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Clifford, Eleanor, Shumailov, Ilia, Zhao, Yiren, Anderson, Ross, Mullins, Robert
Format: Preprint
Published: 2022
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914817296564224
author Clifford, Eleanor
Shumailov, Ilia
Zhao, Yiren
Anderson, Ross
Mullins, Robert
author_facet Clifford, Eleanor
Shumailov, Ilia
Zhao, Yiren
Anderson, Ross
Mullins, Robert
contents Early backdoor attacks against machine learning set off an arms race in attack and defence development. Defences have since appeared demonstrating some ability to detect backdoors in models or even remove them. These defences work by inspecting the training data, the model, or the integrity of the training procedure. In this work, we show that backdoors can be added during compilation, circumventing any safeguards in the data preparation and model training stages. The attacker can not only insert existing weight-based backdoors during compilation, but also a new class of weight-independent backdoors, such as ImpNet. These backdoors are impossible to detect during the training or data preparation processes, because they are not yet present. Next, we demonstrate that some backdoors, including ImpNet, can only be reliably detected at the stage where they are inserted and removing them anywhere else presents a significant challenge. We conclude that ML model security requires assurance of provenance along the entire technical pipeline, including the data, model architecture, compiler, and hardware specification.
format Preprint
id arxiv_https___arxiv_org_abs_2210_00108
institution arXiv
publishDate 2022
record_format arxiv
spellingShingle ImpNet: Imperceptible and blackbox-undetectable backdoors in compiled neural networks
Clifford, Eleanor
Shumailov, Ilia
Zhao, Yiren
Anderson, Ross
Mullins, Robert
Machine Learning
Cryptography and Security
Early backdoor attacks against machine learning set off an arms race in attack and defence development. Defences have since appeared demonstrating some ability to detect backdoors in models or even remove them. These defences work by inspecting the training data, the model, or the integrity of the training procedure. In this work, we show that backdoors can be added during compilation, circumventing any safeguards in the data preparation and model training stages. The attacker can not only insert existing weight-based backdoors during compilation, but also a new class of weight-independent backdoors, such as ImpNet. These backdoors are impossible to detect during the training or data preparation processes, because they are not yet present. Next, we demonstrate that some backdoors, including ImpNet, can only be reliably detected at the stage where they are inserted and removing them anywhere else presents a significant challenge. We conclude that ML model security requires assurance of provenance along the entire technical pipeline, including the data, model architecture, compiler, and hardware specification.
title ImpNet: Imperceptible and blackbox-undetectable backdoors in compiled neural networks
topic Machine Learning
Cryptography and Security
url https://arxiv.org/abs/2210.00108