B1214
Title: Learning mixtures-of-experts from heterogenous and high-dimensional data
Authors: Faicel Chamroukhi - IRT SystemX (France) [presenting]
Abstract: Modern statistical learning algorithms deal with real-world problems involving heterogeneous high-dimensional or functional data. A family of mixture-of-experts models are presented for learning in such situations, as well as their approximation properties and statistical estimation guarantees. These models are considered with high-dimensional predictors or when the predictors are noisy observations from entire functions, and their regularized optimization to provide sparse and interpretable representations.
