Fecha:  1 y 2 de julio de 2014

Ponente: Suresh Manandhar, University of York

Lugar de celebración: Sala 1.03, ETSI Informática, UNED (mapa)

Programa:

  • 1 de julio: Finite mixtures
  • 2 de julio: Dirichlet process infinite mixtures

Resumen:

In this tutorial I will cover the fundamentals of Bayesian mixture
models and show how such models can be applied for various NLP
problems. The tutorial will cover the following topics:

- Beta and Dirichlet distributions
- Dirichlet-Multinomial distribution
- Finite-mixtures
- Gibbs sampling
- Collapsed sampling
- Applications to morphology learning, HMMs
- Latent Dirichlet Allocation
- Applications to word sense induction
- Infinite Mixtures : CRP and Dirichlet processes
- Exchangeability
- Hierarchical Dirichlet process
- Tree-structured DPs and applications to hierarchical clustering

A very minimal background in basic probability theory and machine
learning will be assumed. Students will be expected to have knowledge
of Bayes theorem and its application to computing posterior
distributions.