Encoder-Decoder SNMTF

This repository contains code and experiments accompanying our accepted ICASSP 2026 paper on
Encoder–Decoder Symmetric Nonnegative Matrix Tri-Factorization for graph clustering.

Overview

We introduce an encoder–decoder formulation of symmetric nonnegative matrix tri-factorization (SNMTF) that enforces consistency between graph reconstruction and latent recovery. This coupling yields stable embeddings and well-separated clusters without explicit orthogonality constraints or post-processing.

Requirements

• Python 3.x
• PyTorch
• NumPy
• SciPy
• scikit-learn
• Matplotlib

Usage

Run the main script after placing the dataset in the expected format:

python ED-SNMTF.py

Citation

If you use this code, please cite our ICASSP 2026 paper:

@inproceedings{seyedi2026edsnmtf,
  title     = {Encoder--Decoder Symmetric Nonnegative Matrix Tri-Factorization for Graph Clustering},
  author    = {Seyedi, Amjad and Gillis, Nicolas},
  booktitle = {Proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
  year      = {2026}
}