Denoised Internal Models: a Brain-Inspired Autoencoder against Adversarial Attacks

Abstract

Despite its great success, deep learning severely suffers from robustness; that is, deep neural networks are very vulnerable to adversarial attacks, even the simplest ones. Inspired by recent advances in brain science, we propose the Denoised Internal Models (DIM), a novel generative autoencoder-based model to tackle this challenge. Simulating the pipeline in the human brain for visual signal processing, DIM adopts a two-stage approach. In the first stage, DIM uses a denoiser to reduce the noise and the dimensions of inputs, reflecting the information pre-processing in the thalamus. Inspired from the sparse coding of memory-related traces in the primary visual cortex, the second stage produces a set of internal models, one for each category. We evaluate DIM over 42 adversarial attacks, showing that DIM effectively defenses against all the attacks and outperforms the SOTA on the overall robustness.

arXiv comment16 pages, 3 figures

References in corpus (30)

• Explaining and Harnessing Adversarial Examples
• Deep Learning using Rectified Linear Units (ReLU)
• Ensemble Adversarial Training: Attacks and Defenses
• Deep Speech 2: End-to-End Speech Recognition in English and Mandarin
• XLNet: Generalized Autoregressive Pretraining for Language Understanding
• Towards Deep Learning Models Resistant to Adversarial Attacks
• Wild Patterns: Ten Years After the Rise of Adversarial Machine Learning
• Obfuscated Gradients Give a False Sense of Security: Circumventing Defenses to Adversarial Examples
• Evasion Attacks against Machine Learning at Test Time
• Matrix estimation by Universal Singular Value Thresholding
• Characterizing Adversarial Subspaces Using Local Intrinsic Dimensionality
• Achieving Human Parity in Conversational Speech Recognition
• Countering Adversarial Images using Input Transformations
• Benchmarking Neural Network Robustness to Common Corruptions and Perturbations
• Defense-GAN: Protecting Classifiers Against Adversarial Attacks Using Generative Models
• Towards the first adversarially robust neural network model on MNIST
• Stochastic Activation Pruning for Robust Adversarial Defense
• Mitigating Adversarial Effects Through Randomization
• Improving Adversarial Robustness via Promoting Ensemble Diversity
• Foolbox: A Python toolbox to benchmark the robustness of machine learning models
• On Adaptive Attacks to Adversarial Example Defenses
• Decision-Based Adversarial Attacks: Reliable Attacks Against Black-Box Machine Learning Models
• On the Robustness of the CVPR 2018 White-Box Adversarial Example Defenses
• Rethinking Softmax Cross-Entropy Loss for Adversarial Robustness
• Accurate, reliable and fast robustness evaluation
• A New Defense Against Adversarial Images: Turning a Weakness into a Strength
• ME-Net: Towards Effective Adversarial Robustness with Matrix Estimation
• GAT: Generative Adversarial Training for Adversarial Example Detection and Robust Classification
• On the Need for Topology-Aware Generative Models for Manifold-Based Defenses
• Fast Differentiable Clipping-Aware Normalization and Rescaling