Unsupervised Selective Manifold Regularized Matrix Factorization

TL;DR

This paper introduces an unsupervised, selective manifold regularized matrix factorization method that improves data representation by learning sparse representatives and their affinities, outperforming existing algorithms.

Contribution

It proposes a novel selective regularization approach that jointly learns representatives, affinities, and factorization, with a fast approximation for efficiency.

Findings

Competitive performance against state-of-the-art methods

Effective in preserving neighborhood structures

Improves data factorization quality

Abstract

Manifold regularization methods for matrix factorization rely on the cluster assumption, whereby the neighborhood structure of data in the input space is preserved in the factorization space. We argue that using the k-neighborhoods of all data points as regularization constraints can negatively affect the quality of the factorization, and propose an unsupervised and selective regularized matrix factorization algorithm to tackle this problem. Our approach jointly learns a sparse set of representatives and their neighbor affinities, and the data factorization. We further propose a fast approximation of our approach by relaxing the selectivity constraints on the data. Our proposed algorithms are competitive against baselines and state-of-the-art manifold regularization and clustering algorithms.