BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:What's the right complexity measure for inferring causal relations
 ? - Dominik Janzing (Max Planck Institute Tuebingen)
DTSTART:20100226T153000Z
DTEND:20100226T163000Z
UID:TALK22581@talks.cam.ac.uk
CONTACT:8047
DESCRIPTION:If X causes Y for two random variables X and Y\, we expect tha
 t the\nfactorization of P(X\,Y) into P(X)P(Y|X) is simpler than the *non-c
 ausal*\nfactorization P(Y)P(X|Y). This is because P(Y|X) describes the cau
 sal\nmechanism while P(X|Y) is "only a mathematical expression".\n\nDiscus
 sions have shown that a lot of researchers agree on this\nintuition. Since
  we would like to use this principle for inferring\ncausal directions\, we
  are left with two problems:\n\n(1) what does "simple" mean?\n\n(2) is the
 re any deeper justification for this principle?\n\nOur answer to question 
 (2) is a clear "yes" if complexity vs simplicity\nis measured in terms of 
 Kolmogorov complexity: I will present a theory\nof causal inference that g
 eneralizes the framework of Bayesian networks\nto *algorithmic* instead of
  *statistical* conditional dependences. I\nwill show that our theory impli
 es the above inference principle.\nHowever\, since Kolmogorov complexity i
 s uncomputable we still need\ncomplexity measures that are appropriate for
  practical implementations.\n\nI will present some first small steps towar
 ds this challenging goal.\n\n\n\n\nhttp://www.kyb.mpg.de/~janzing
LOCATION:MR12\, CMS\, Wilberforce Road\, Cambridge\, CB3 0WB
END:VEVENT
END:VCALENDAR