Tid: 19 januari 2009 kl 15.15-17.00
Plats : Seminarierummet 3733, Institutionen för matematik, KTH, Lindstedts väg 25, plan 7. Karta!
Föredragshållare: Professor Stefan Arnborg, NADA, KTH
Titel: The legacy of Ed Jaynes
Sammanfattning: Ed Jaynes (1922-1998) was a colorful proponent of Bayesian and Maximum Entropy methods in applied data analysis. His lecture notes 'Probability Theory: The Logic of Science' were published by Cambridge University Press in 2003, but had worked as a powerful Samizdat for Bayesian underground fanatics for a decade. It is sometimes thought that his (and Lindley's) arrogant writing polarized the community unnecessarily and delayed the powerful syntheses advocated by, e.g., Andrew Gelman and his colleagues. But the main danger statistics runs is that it can become boring, and a good fight can thus have a direct economic value for science.
I will go through some threads in Jaynes' work and see how they were approached by later investigators. Of particular interest is his strong belief in the unconditional validity of Cox's logic-based 'derivation' of the inevitability of Bayesian analysis (a parallell to Lindley's 1982 paper), and also his rather fundamentalist approach to maximum entropy methods as more justified than the alternatives, often based on Dempster's, Shafer's or Walley's work. Some of the results I present is joint work with Gunnar Sjödin and Joel Brynielsson.
|Sidansvarig: Filip Lindskog