Clustering is one of the most important task in Machine Learning. It consists in grouping together similar objects and is often considered to be an information compression task. In real life applications, clustering is often applied to online data to process live stream that may come from several sources.

In this work, we propose an incremental version of collaborative clustering based on information theory. Our algorithm has the advantage of handling incremental learning while combining the strengths of both information theory methods that have proved to be efficient and ideal candidates for clustering tasks, and collaborative clustering algorithms that have the capacity to process data from several sources such as neural network. Our proposed model relies on a Kolmogorov complexity based formulation of collaborative clustering, and data-wise division and optimization of this objective function, thus enabling to individually process multiview data as they arrive. The proposed incremental method also has a lower computational complexity than the original batch version.

Our approach was tested on several datasets and compared with other state of the art methods and has proved to be very promising.