Aristides Gionis, HIIT/BRU
Learning a mixture of models is a tool often employed in order to understand the hidden structure and the characteristics of a data set. A resulting mixture model can be used for prediction, classification, summarization, and visualization of data.
An alternative approach to understanding the structure of data is via the formulation of clustering. Modeling of data using mixture models is commonly used in statistics, AI, and machine learning, while clustering is more typically studied in theoretical computer science, databases, and data mining.
In this talk, I will attempt to investigate the connections between learning mixture models and clustering. In addition, I will describe a simple combinatorial algorithm that provides approximation guarantees to the problem of learning a mixture model.
This is joint work with Heikki Mannila and Taneli Mielikäinen.