An Introduction to Bayesian Methods for Interaction Design

Abstract

Bayesian modelling has much to offer those working in human-computer interaction, but many of the concepts are alien. This chapter introduces Bayesian modelling in interaction design. The chapter outlines the philosophical stance that sets Bayesian approaches apart, as well as a light introduction to the nomenclature and computational and mathematical machinery. We discuss specific models of relevance to interaction, including probabilistic filtering, nonparametric Bayesian inference, approximate Bayesian computation and belief networks. We include a worked example of a Fitts' law modelling task from a Bayesian perspective, applying Bayesian linear regression via a probabilistic program. We identify five distinct facets of Bayesian interaction: probabilistic interaction in the control loop; Bayesian optimisation at design time; analysis of empirical results with Bayesian statistics; visualisation and interaction with Bayesian models; and Bayesian cognitive modelling of users. We conclude with a discussion of the pros and cons of Bayesian approaches, the ethical implications therein and suggestions for further reading.