riemannian-streaming-cca

This repository contains the reference code for our paper An Online Riemannian PCA for Stochastic Canonical Correlation Analysis (NeurIPS-2021)

Requirements

For CCA experiments on MNIST, CIFAR10 and Mediamill:

• Matlab
• manopt for Matlab

For DeepCCA experiments:

• Python 3
• Pytorch 1.5+
• sklearn

CCA experiments

Download the mat files for MNIST, CIFAR10 and Mediamill from google drive here and put them in ./CCA

Use Matlab to run demo_run.m

DeepCCA experiments

cd DeepCCA

First use download_data.sh to download the data (we utilize the data loader file and part of the scripts from this DeepCCA repo)

To run the DeepCCA experiments:

python main.py --feat_dim=100

where --feat_dim specifies the dimension of the output feature by DNN.

Reference

If you find our work useful, please consider citing our paper.

@article{meng2021online,
  title={An Online Riemannian PCA for Stochastic Canonical Correlation Analysis},
  author={Meng, Zihang and Chakraborty, Rudrasis and Singh, Vikas},
  journal={arXiv preprint arXiv:2106.07479},
  year={2021}
}