Bypassing Backdoor Detection Algorithms in Deep Learning
This work highlights a vulnerability in current backdoor detection algorithms, calling for adversary-aware defenses, but it is incremental as it builds on existing adversarial training methods.
The paper tackles the problem of backdoor detection in deep learning by designing an adversarial backdoor embedding algorithm that bypasses existing detection methods, including state-of-the-art techniques, by optimizing model loss and maximizing indistinguishability in hidden representations.
Deep learning models are vulnerable to various adversarial manipulations of their training data, parameters, and input sample. In particular, an adversary can modify the training data and model parameters to embed backdoors into the model, so the model behaves according to the adversary's objective if the input contains the backdoor features, referred to as the backdoor trigger (e.g., a stamp on an image). The poisoned model's behavior on clean data, however, remains unchanged. Many detection algorithms are designed to detect backdoors on input samples or model parameters, through the statistical difference between the latent representations of adversarial and clean input samples in the poisoned model. In this paper, we design an adversarial backdoor embedding algorithm that can bypass the existing detection algorithms including the state-of-the-art techniques. We design an adaptive adversarial training algorithm that optimizes the original loss function of the model, and also maximizes the indistinguishability of the hidden representations of poisoned data and clean data. This work calls for designing adversary-aware defense mechanisms for backdoor detection.