|Stochastic Systems Group|
Bayesian Hypothesis Testing with Prototype Priors
Kush R. Varshney
A bounded rationality model of Bayesian hypothesis testing is developed through quantization of prior probabilities of the hypotheses taken as random variables. Nearest neighbor and centroid conditions for optimal quantization are derived using mean Bayes risk error as a distortion measure. The implications of the decision theory analysis are discussed in terms of information-based discrimination. A generative model for own-race bias in decision-making is provided. Properties of the utility function required for empirical studies of discrimination (such as by police, human resources, and sports referees) to match model predictions are noted.
Joint work with Lav Varshney.
Problems with this site should be emailed to firstname.lastname@example.org