Robust Training Using Natural Transformation
arXiv:2105.04070
Abstract
Previous robustness approaches for deep learning models such as data augmentation techniques via data transformation or adversarial training cannot capture real-world variations that preserve the semantics of the input, such as a change in lighting conditions. To bridge this gap, we present NaTra, an adversarial training scheme that is designed to improve the robustness of image classification algorithms. We target attributes of the input images that are independent of the class identification, and manipulate those attributes to mimic real-world natural transformations (NaTra) of the inputs, which are then used to augment the training dataset of the image classifier. Specifically, we apply \textit{Batch Inverse Encoding and Shifting} to map a batch of given images to corresponding disentangled latent codes of well-trained generative models. \textit{Latent Codes Expansion} is used to boost image reconstruction quality through the incorporation of extended feature maps. \textit{Unsupervised Attribute Directing and Manipulation} enables identification of the latent directions that correspond to specific attribute changes, and then produce interpretable manipulations of those attributes, thereby generating natural transformations to the input data. We demonstrate the efficacy of our scheme by utilizing the disentangled latent representations derived from well-trained GANs to mimic transformations of an image that are similar to real-world natural variations (such as lighting conditions or hairstyle), and train models to be invariant to these natural transformations. Extensive experiments show that our method improves generalization of classification models and increases its robustness to various real-world distortions
arXiv commentarXiv admin note: text overlap with arXiv:1912.03192, arXiv:2004.02546 by other authors
References in corpus (17)
- Very Deep Convolutional Networks for Large-Scale Image Recognition
- mixup: Beyond Empirical Risk Minimization
- Large Scale GAN Training for High Fidelity Natural Image Synthesis
- Conditional Image Synthesis With Auxiliary Classifier GANs
- Towards Deep Learning Models Resistant to Adversarial Attacks
- LSUN: Construction of a Large-scale Image Dataset using Deep Learning with Humans in the Loop
- AutoAugment: Learning Augmentation Policies from Data
- Invertible Conditional GANs for image editing
- GANSpace: Discovering Interpretable GAN Controls
- Generalisation in humans and deep neural networks
- Generating Adversarial Examples with Adversarial Networks
- Adversarial Transformation Networks: Learning to Generate Adversarial Examples
- GAN Dissection: Visualizing and Understanding Generative Adversarial Networks
- Examining the Impact of Blur on Recognition by Convolutional Networks
- The Robust Manifold Defense: Adversarial Training using Generative Models
- Differentiable Augmentation for Data-Efficient GAN Training
- Precise Recovery of Latent Vectors from Generative Adversarial Networks