aradha / recursive_feature_machines

Recursive Feature Machines

There are two notebooks to test out RFM:

• low_rank.ipynb (an example of low rank polynomials)
• svhn.ipynb (for the SVHN dataset)

Requirements:

1. Python 3
2. pytorch==1.13.0
3. torchvision==0.14.0
4. hickle==5.0.2
5. tqdm==4.64.1

Paper

Feature learning in neural networks and kernel machines that recursively learn features