This is the repo for the Learning and Aligning Structured Random Feature Networks paper, accepted at the ICLR Re-Align 2024 workshop. The code for our factorized covariance and rainbow sampling methods can be found in the FactConv directory and the code for Randomized Scattering Networks can be found in RSN_experiments.
Our Factorized Covariance module is located in FactConv/conv_modules.py and can be called similarly to a nn.Conv2d module, like so m = FactConv2d(in_channels=3, out_channels=32, kernel_size=(3,3)).
V1 initalization based on the receptive V1 field of mice is located in FactConv/V1_covariance.py and can be called like so; V1_init(m, size=2, spatial_freq=0.1, scale=1, center=center) where center=((m.kernel_size[0]-1)/2, (m.kernel_size[1]-1)/2).
Run python3 setup.py install to install the Factored Covariance module
Rainbow sampling can be done with our RainbowSampler class object as so:
from rainbow import RainbowSampler
R = RainbowSampler(net, trainloader)
rainbow_net = R.sample()
To use our factorized ResNet in our rainbow sampling procedure as outlined in "Learning and Aligning Structured Random Feature Networks" by White et al., specify RainbowSampler(..., sampling='structured_alignment', wa=True, in_wa=True, aca=True). This can be specified for both FactConv2d and nn.Conv2d modules.
To do the rainbow sampling procedure of "A Rainbow in Deep Network Black Boxes" by Guth et al., specify RainbowSampler(..., sampling='cc_specification', wa=False, aca=True). This is specified specifically for networks using nn.Conv2d modules.
This google drive link contains our pretrained ResNet18 networks, with both convolutional and factorized convolutional varations.
If you found this repository or our paper helpful, please cite us as shown below:
@inproceedings{
white2024learning,
title={Learning and Aligning Structured Random Feature Networks},
author={Vivian White and Muawiz Sajjad Chaudhary and Guy Wolf and Guillaume Lajoie and Kameron Decker Harris},
booktitle={ICLR 2024 Workshop on Representational Alignment},
year={2024},
url={https://openreview.net/forum?id=vWhUQXQoFF}
}