# Neural Networks with Recurrent Generative Feedback

arXiv: https://arxiv.org/abs/2007.09200 · HTML: https://mlsanity.com/paper/2007.09200/

## Key facts

- Cited by 23 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 17 Jul 2020.

## Built on by (most cited first, 23 of 23)

- [Test-Time Prompt Tuning for Zero-Shot Generalization in Vision-Language Models](https://mlsanity.com/paper/2209.07511/) (2022)
- [MEMO: Test Time Robustness via Adaptation and Augmentation](https://mlsanity.com/paper/2110.09506/) (2021)
- [Your Diffusion Model is Secretly a Zero-Shot Classifier](https://mlsanity.com/paper/2303.16203/) (2023)
- [Semantic Segmentation with Generative Models: Semi-Supervised Learning and Strong Out-of-Domain Generalization](https://mlsanity.com/paper/2104.05833/) (2021)
- [Improving Transformers with Probabilistic Attention Keys](https://mlsanity.com/paper/2110.08678/) (2021)
- [Hybrid Predictive Coding: Inferring, Fast and Slow](https://mlsanity.com/paper/2204.02169/) (2022)
- [Predify: Augmenting deep neural networks with brain-inspired predictive coding dynamics](https://mlsanity.com/paper/2106.02749/) (2021)
- [Top-Down Visual Attention from Analysis by Synthesis](https://mlsanity.com/paper/2303.13043/) (2023)
- [Temporal Coherent Test-Time Optimization for Robust Video Classification](https://mlsanity.com/paper/2302.14309/) (2023)
- [Predictive coding feedback results in perceived illusory contours in a recurrent neural network](https://mlsanity.com/paper/2102.01955/) (2021)
- [CyberV: Cybernetics for Test-time Scaling in Video Understanding](https://mlsanity.com/paper/2506.07971/) (2025)
- [Symmetrical Flow Matching: Unified Image Generation, Segmentation, and Classification with Score-Based Generative Models](https://mlsanity.com/paper/2506.10634/) (2025)
- [Augmenting BERT-style Models with Predictive Coding to Improve Discourse-level Representations](https://mlsanity.com/paper/2109.04602/) (2021)
- [Human Eyes Inspired Recurrent Neural Networks are More Robust Against Adversarial Noises](https://mlsanity.com/paper/2206.07282/) (2022)
- [Visual Attention Emerges from Recurrent Sparse Reconstruction](https://mlsanity.com/paper/2204.10962/) (2022)
- [Empirical Advocacy of Bio-inspired Models for Robust Image Recognition](https://mlsanity.com/paper/2205.09037/) (2022)
- [On the role of feedback in visual processing: a predictive coding perspective](https://mlsanity.com/paper/2106.04225/) (2021)
- [Classification and Reconstruction Processes in Deep Predictive Coding Networks: Antagonists or Allies?](https://mlsanity.com/paper/2401.09237/) (2024)
- [MedSymmFlow: Bridging Generative Modeling and Classification in Medical Imaging through Symmetrical Flow Matching](https://mlsanity.com/paper/2507.19098/) (2025)
- [Enhancing Audio-Language Models through Self-Supervised Post-Training with Text-Audio Pairs](https://mlsanity.com/paper/2408.09269/) (2024)
- [Contextual Feedback Loops: Amplifying Deep Reasoning with Iterative Top-Down Feedback](https://mlsanity.com/paper/2412.17737/) (2024)
- [Quasi-Equivariant Metanetworks](https://mlsanity.com/paper/2604.23720/) (2026)
- [Feedback RoI Features Improve Aerial Object Detection](https://mlsanity.com/paper/2311.17129/) (2023)

## Abstract (arXiv)

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.