|
Abstract:
|
In many high-dimensional structured input/output problems, where both input and output can contain tens of thousands, sometimes even millions of inter-related features, learning a sparse and consistent structured predictive function can be of paramount importance for both robustness and interpretability of the model. This talk will present a new statistical formalism known as the maximum entropy discrimination Markov networks, which address the problem of estimating sparse structured I/O models under a maximum margin framework, but using a entropic regularizer that lead to a distribution of structured prediction functions that are simultaneously primal and dual sparse (i.e., with few support vectors, and of low effective feature dimension), and can be efficiently solved via a novel algorithm that builds on variational inference and existing solvers for the maximum margin Markov network.
|
