Neural Networks with Recurrent Generative Feedback

Abstract

Neural networks are vulnerable to input perturbations such as additive noise and adversarial attacks. In contrast, human perception is much more robust to such perturbations. The Bayesian brain hypothesis states that human brains use an internal generative model to update the posterior beliefs of the sensory input. This mechanism can be interpreted as a form of self-consistency between the maximum a posteriori (MAP) estimation of an internal generative model and the external environment. Inspired by such hypothesis, we enforce self-consistency in neural networks by incorporating generative recurrent feedback. We instantiate this design on convolutional neural networks (CNNs). The proposed framework, termed Convolutional Neural Networks with Feedback (CNN-F), introduces a generative feedback with latent variables to existing CNN architectures, where consistent predictions are made through alternating MAP inference under a Bayesian framework. In the experiments, CNN-F shows considerably improved adversarial robustness over conventional feedforward CNNs on standard benchmarks.

arXiv commentNeurIPS 2020

References in corpus (18)

• Adam: A Method for Stochastic Optimization
• Very Deep Convolutional Networks for Large-Scale Image Recognition
• Instance Normalization: The Missing Ingredient for Fast Stylization
• Wide Residual Networks
• Towards Deep Learning Models Resistant to Adversarial Attacks
• Rethinking the Inception Architecture for Computer Vision
• Recurrence is required to capture the representational dynamics of the human visual system
• Adversarial Examples that Fool both Computer Vision and Time-Limited Humans
• Task-Driven Convolutional Recurrent Models of the Visual System
• The Consciousness Prior
• Recurrent Independent Mechanisms
• Deep Predictive Coding Network for Object Recognition
• A Theoretical Framework for Target Propagation
• Learning to Combine Top-Down and Bottom-Up Signals in Recurrent Neural Networks with Attention over Modules
• Unsupervised Learning from Continuous Video in a Scalable Predictive Recurrent Network
• Multi-Object Classification and Unsupervised Scene Understanding Using Deep Learning Features and Latent Tree Probabilistic Models
• A Bayesian Perspective of Convolutional Neural Networks through a Deconvolutional Generative Model
• State-Reification Networks: Improving Generalization by Modeling the Distribution of Hidden Representations

Cited by in corpus (3)

• MEMO: Test Time Robustness via Adaptation and Augmentation
• Predify: Augmenting deep neural networks with brain-inspired predictive coding dynamics
• Predictive coding feedback results in perceived illusory contours in a recurrent neural network