Achieving Robustness in the Wild via Adversarial Mixing with Disentangled Representations
This addresses robustness issues in computer vision models for real-world applications, offering a novel approach beyond traditional adversarial training, though it is incremental in leveraging existing disentanglement techniques.
The paper tackles the problem of deep learning models failing to generalize to real-world variations like lighting changes by proposing a method using disentangled representations from StyleGAN to generate adversarial perturbations that preserve semantics, resulting in improved generalization and reduced spurious correlations, such as a 21% error reduction for a smile detector.
Recent research has made the surprising finding that state-of-the-art deep learning models sometimes fail to generalize to small variations of the input. Adversarial training has been shown to be an effective approach to overcome this problem. However, its application has been limited to enforcing invariance to analytically defined transformations like $\ell_p$-norm bounded perturbations. Such perturbations do not necessarily cover plausible real-world variations that preserve the semantics of the input (such as a change in lighting conditions). In this paper, we propose a novel approach to express and formalize robustness to these kinds of real-world transformations of the input. The two key ideas underlying our formulation are (1) leveraging disentangled representations of the input to define different factors of variations, and (2) generating new input images by adversarially composing the representations of different images. We use a StyleGAN model to demonstrate the efficacy of this framework. Specifically, we leverage the disentangled latent representations computed by a StyleGAN model to generate perturbations of an image that are similar to real-world variations (like adding make-up, or changing the skin-tone of a person) and train models to be invariant to these perturbations. Extensive experiments show that our method improves generalization and reduces the effect of spurious correlations (reducing the error rate of a "smile" detector by 21% for example).