Tid: 30 oktober 2000 kl 1515-1700
Plats : Seminarierummet 3733, Institutionen för matematik, KTH, Lindstedts väg 25, plan 7. Karta!
Föredragshållare: Gunnar Englund, Matematisk statistik, KTH.
Titel: Markov Chain Monte Carlo, contingency tables and Gröbner bases
Sammanfattning: Markov Chain Monte Carlo (MCMC) is a powerful technique to simulate from complicated distributions which has been used extensively in e g Bayesian analysis. MCMC can also be used to analyse e.g. contingency tables and logistic regression or more generally conditional distributions given restrictions (sufficient statistics) by simulation without having to resort to asymptotic -distributions. A simple Markov chain is constructed which has the desired distribution as a stationary distribution. Gröbner bases can be used to ensure that the resulting Markov chain is irreducible and aperiodic and hence ergodic.
The talk is based on the article "Algebraic algorithms for sampling from conditional distributions" by Persi Diaconis and Bernd Sturmfels in Annals of Statistics 1998 (Vol. 26).
Preliminary version of overheads (English) or Swedish.
To the list of seminars