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

arXiv:2210.00108v4 Announce Type: replace
Abstract: 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 …

abstract arxiv attacks backdoor blackbox cs.cr cs.lg data defence development integrity machine machine learning networks neural networks race set them training training data type work