COPER: Correlation-based Permutations for Multi-View Clustering (ICLR 2025)

Official implementation of the accepted paper.

We propose an end-to-end deep learning-based multi-view clustering framework for general data types (such as images and tables). Our approach involves generating meaningful fused representations using a novel permutation-based canonical correlation objective. Cluster assignments are learned by identifying consistent pseudo-labels across multiple views.

How to run:

0. Download the datasets here: OneDrive, Zenodo
   • please see the datasets details in DATASETS.md
1. Update the path to the dataset in config files conf/<dataset_name>.yaml
2. Run training: python coper.py --cfg <config file>
3. Please note that for improved stability the model is trained with decoders by default.

Authors:

• Jonathan Svirsky
• Ran Eisenberg
• Ofir Lindenbaum

Citing:

If you are using this code or datasets, please cite our paper:

@inproceedings{
eisenberg2025coper,
title={{COPER}: Correlation-based Permutations for Multi-View Clustering},
author={Ran Eisenberg and Jonathan Svirsky and Ofir Lindenbaum},
booktitle={The Thirteenth International Conference on Learning Representations},
year={2025},
url={https://openreview.net/forum?id=5ZEbpBYGwH}
}