CorrGAN: Input Transformation Technique Against Natural Corruptions
This addresses the problem of improving DNN robustness against natural corruptions for real-time systems like autonomous vehicles, though it is incremental as it builds on existing GAN frameworks.
The paper tackles the vulnerability of Deep Neural Networks (DNNs) to natural corruptions like fog and blur in autonomous vehicles by proposing CorrGAN, a method that denoises corrupted inputs to generate benign versions, resulting in up to 75.2% of previously misclassified inputs being correctly classified.
Because of the increasing accuracy of Deep Neural Networks (DNNs) on different tasks, a lot of real times systems are utilizing DNNs. These DNNs are vulnerable to adversarial perturbations and corruptions. Specifically, natural corruptions like fog, blur, contrast etc can affect the prediction of DNN in an autonomous vehicle. In real time, these corruptions are needed to be detected and also the corrupted inputs are needed to be de-noised to be predicted correctly. In this work, we propose CorrGAN approach, which can generate benign input when a corrupted input is provided. In this framework, we train Generative Adversarial Network (GAN) with novel intermediate output-based loss function. The GAN can denoise the corrupted input and generate benign input. Through experimentation, we show that up to 75.2% of the corrupted misclassified inputs can be classified correctly by DNN using CorrGAN.