Unsupervised Selective Manifold Regularized Matrix Factorization
This addresses a specific issue in manifold regularization for matrix factorization, offering an incremental improvement for researchers in unsupervised learning and dimensionality reduction.
The paper tackles the problem that using all data point neighborhoods in manifold-regularized matrix factorization can degrade factorization quality, and proposes an unsupervised selective algorithm that jointly learns sparse representatives and neighbor affinities, achieving competitive performance against baselines and state-of-the-art methods.
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.