Feedback feedforward alignment

This repository is the official implementatin of "Brain-like Flexible Visual Inference by Harnessing Feedback Feedforward Alignment" appeared in NeurIPS 2023.

**We are adding more code to this repo **

Learning phase (Training)

The checkpoints are available at the following links:

	FFA	BP	FA	checkpoints
MNIST	discr:98.87%, recons:0.98	discr:99.47%, recons:0.02	discr:97.51%, recons:0.01	download
CIFAR10	discr:76.47%, recons:0.88	discr:88.53%, recons:0.01	discr:71.56%, recons:0.01	download

Validation

To generate figures for training evaluation FeedbackFeedforwradAlignment/Generate_Figures/Fig2_Training_results.ipynb To generate figures for robustness evaluation

FeedbackFeedforwradAlignment/Generate_Figures/Fig3_robustness_evaluation.ipynb

1. Accuracy

2. Reconstruction performance

3. Alignment

4. Noise robustness

5. Adversarial robustness

Inference phase (brain-like flexible visual inference such as de-occlusions, imagination, and hallucinations)

Citation

@inproceedings{ Toosi2023brainlike,

title={Brain-like Flexible Visual Inference by Harnessing Feedback Feedforward Alignment},

author={Toosi, Tahereh and Issa, Elias B},

booktitle={Thirty-seventh Conference on Neural Information Processing Systems},

year={2023},

url={https://openreview.net/forum?id=DBlkX8Nczr}

}