Unsupervised spectral learning

TL;DR

This paper introduces an unsupervised spectral clustering method that learns the similarity function iteratively from observed features while simultaneously performing data clustering, demonstrating promising results on synthetic and real datasets.

Contribution

It presents a novel algorithm that performs spectral clustering without pre-defined similarity matrices by learning the similarity function during clustering.

Findings

Effective on synthetic data

Promising results on real data

Simultaneous learning and clustering

Abstract

In spectral clustering and spectral image segmentation, the data is partioned starting from a given matrix of pairwise similarities S. the matrix S is constructed by hand, or learned on a separate training set. In this paper we show how to achieve spectral clustering in unsupervised mode. Our algorithm starts with a set of observed pairwise features, which are possible components of an unknown, parametric similarity function. This function is learned iteratively, at the same time as the clustering of the data. The algorithm shows promosing results on synthetic and real data.