Lawrence Joseph: Bayesian Inference in Music

ABSTRACT

The two main paradigms for statistical inference have been classical (or frequentist) analysis, leading to the familiar p-values and confidence intervals, and Bayesian analysis, leading to posterior densities. While much debate has focused on philosophical differences between the paradigms, in this talk I will concentrate on the practical differences.  I will compare and contrast the inferences available from both paradigms through a series of examples from musical experiments, pointing out where Bayesian analysis may be most advantageous. Examples would include better designs for musical experiments across all domains based on Bayesian principles, and furthering work on cognition based musical structures, similar to that done by David Temperley. 

 

ABOUT LAWRENCE JOSEPH

Lawrence Joseph obtained a PhD in Mathematical Statistics from McGill University in 1990. Since that time, he has been a professor in Department of Epidemiology and Biostatistics in the Faculty of Medicine at McGill. His main interest has been in Bayesian experimental design and analysis. In particular, he has developed Bayesian models for diagnostic testing data, change point problems, and published new Bayesian criteria for experimental design including sample size methods.  He has always had a strong interest in music, with activities ranging from programming in musical languages such as pd and CSound, building home made analog musical instruments, playing guitar, and writing about avant-garde jazz and free improvisation for the Montreal Mirror, Signal to Noise and other publications.