Adversarial Images for Variational Autoencoders

Abstract

We investigate adversarial attacks for autoencoders. We propose a procedure that distorts the input image to mislead the autoencoder in reconstructing a completely different target image. We attack the internal latent representations, attempting to make the adversarial input produce an internal representation as similar as possible as the target's. We find that autoencoders are much more robust to the attack than classifiers: while some examples have tolerably small input distortion, and reasonable similarity to the target image, there is a quasi-linear trade-off between those aims. We report results on MNIST and SVHN datasets, and also test regular deterministic autoencoders, reaching similar conclusions in all cases. Finally, we show that the usual adversarial attack for classifiers, while being much easier, also presents a direct proportion between distortion on the input, and misdirection on the output. That proportionality however is hidden by the normalization of the output, which maps a linear layer into non-linear probabilities.

Workshop on Adversarial Training, NIPS 2016, Barcelona, Spain

References in corpus (6)

• Auto-Encoding Variational Bayes
• Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks
• Stochastic Backpropagation and Approximate Inference in Deep Generative Models
• DRAW: A Recurrent Neural Network For Image Generation
• Towards Deep Neural Network Architectures Robust to Adversarial Examples
• Attend, Infer, Repeat: Fast Scene Understanding with Generative Models

Cited by in corpus (19)

• Adversarial Examples: Attacks and Defenses for Deep Learning
• CycleGAN, a Master of Steganography
• Adversarial Attacks Against Deep Generative Models on Data: A Survey
• Blocking Transferability of Adversarial Examples in Black-Box Learning Systems
• Do Deep Generative Models Know What They Don't Know?
• A Survey on State-of-the-art Deep Learning Applications and Challenges
• Towards Robust Neural Image Compression: Adversarial Attack and Model Finetuning
• When and How to Fool Explainable Models (and Humans) with Adversarial Examples
• Are Generative Classifiers More Robust to Adversarial Attacks?
• Adversarial Attacks on Variational Autoencoders
• Say What I Want: Towards the Dark Side of Neural Dialogue Models
• Representation Learning with Autoencoders for Electronic Health Records: A Comparative Study
• Category-wise Attack: Transferable Adversarial Examples for Anchor Free Object Detection
• Protecting Against Image Translation Deepfakes by Leaking Universal Perturbations from Black-Box Neural Networks
• Interpreting Adversarial Examples with Attributes
• Improving VAEs' Robustness to Adversarial Attack
• Towards a Theoretical Understanding of the Robustness of Variational Autoencoders
• Certifiably Robust Variational Autoencoders
• Fast Local Attack: Generating Local Adversarial Examples for Object Detectors