Iterative Spectral Method for Alternative Clustering

It is extensively used for exploratory data analysis. Traditional clustering algorithms typically identify a single partitioning of a given dataset. However, data is often multifaceted and can be both interpreted and clustered through multiple viewpoints (or, views). For example, the same face data can be clustered based on either identity or based on pose. In real applications, partitions generated by a clustering algorithm may not correspond to the view a user is interested in. In this paper, we address the problem of finding an alternative clustering, given a dataset and an existing, pre-computed clustering. Ideally, one would like the alternative clustering to be novel (i.e., non-redundant) w.r.t. the existing clustering to reveal a new viewpoint to the user. Simultaneously, one would like the result to reveal partitions of high clustering quality. Several recent papers propose algorithms for alternativeProceedings of the 21 st International Conference on Artificial Intelligence and Statistics (AISTATS) 2018, Lan-zarote, Spain.