ARLIMVC: Adaptive Representation Learning Framework for Incomplete Multi-View Clustering

Intro

This repository contains the code of our paper Adaptive Representation Learning Framework for Incomplete Multi-View Clustering (ARLIMVC).

ARLIMVC is a deep incomplete multi-view clustering framework built on a variational autoencoder architecture. It integrates three key components: a mutual information-guided data recovery module, a dual-constraint representation alignment module, and a KAN-based weighted representation fusion module to learn robust and discriminative representations from incomplete multi-view data.

Adaptive Representation Learning Framework for Incomplete Multi-View Clustering

Highlights

• Mutual information-guided data recovery for adaptive missing-view reconstruction.
• Dual-constraint representation alignment to enhance both cross-view consistency and intra-view compactness.
• KAN-based weighted fusion to dynamically capture the contribution of different views.
• End-to-end optimization with VAE-based representation learning and clustering refinement.

Datasets

The datasets used in our paper include COIL20, 100leaves, Handwritten, MSRC, ORL, Scene-15, and ALOI-100.
In the current public implementation, we provide the default configuration for COIL20.

Requirements

Recommended environment:

• Python 3.8+
• PyTorch
• NumPy
• SciPy
• scikit-learn
• matplotlib
• seaborn
• hnswlib
• munkres

You can install the main dependencies with:

pip install torch numpy scipy scikit-learn matplotlib seaborn hnswlib munkres