A framework for deep constrained clustering

The area of constrained clustering has been extensively explored by researchers and used practitioners. Constrained formulations exist for popular algorithms such as k-means, mixture models, spectral but have several limitations. A fundamental strength deep learning is its flexibility, here we explore a framework in particular how it can extend the field clustering. We show that our not only handle standard together/apart constraints (without well documented negative effects reported earlier) generated from labeled side information more complex new types continuous values high-level domain knowledge. Furthermore, propose an efficient training paradigm generally applicable to these four constraints. validate effectiveness approach empirical results on both image text datasets. also study robustness when with noisy different components contribute final performance. Our source code available at: http://github.com/blueocean92 .