Features of Bayesian Learning Based on Conditioning Using Conditional Expectations

This abstract has open access
Abstract Summary

Miklos Redei (London School of Economics), Zalan Gyenis (Jagiellonian University and Eotvos University)

Conditional expectations define a "Bayes accessibility" relation among probability measures on a Boolean algebra. If a probability measure is Bayes accessible from another, then the Bayesian Agent can learn this probability from the evidence represented by the other probability. The Bayes Blind Spot is the set of probability measures on a Boolean algebra that cannot be learned by a single conditionalization from any evidence. It is shown that the Bayes Blind Spot is uncountably infinite in standard probability spaces and that it is a large set (in cardinality, in measure and topologically) if the Boolean algebra is finite.

Submission ID :
NKDR712
Abstract Topics
Jagiellonian University and Eotvos University
London School of Economics
399 visits