K-means algorithm with Bregman divergences and constructing predictive models based on this algorithm
schedule le lundi 04 mars 2019 de 17h00 à 18h00
Organisé par : G. Conchon-Kerjan, F. Coppini, B. Dembin
Intervenant : Sothea Has (LPSM)
Lieu : salle 1016 au bâtiment Sophie Germain à Paris 7 Diderot
Sujet : K-means algorithm with Bregman divergences and constructing predictive models based on this algorithm
When we take about supervised learning problems, we are interested in explaining the values of some response variable or output $Y$ using some predictors or input $X$. On the other hand, the goal of an unsupervised learning problem is to understand the structure or the relationship between variables of the input data itself. In this talk, we will introduce the definition of Bregman divergences which are the members of a broad class of dissimilarity measures and its relationship with a famous Exponential family. This relationship is a strong motivation of using this tool in an unsupervised learning method known as K-means clustering algorithm with Bregman divergences to approximate the group structure of the input data. Then, we will briefly describe a way to construct predictive models in supervised learning problems partially based on this unsupervised learning method. The numerical results performed on some simulated data are also provided.