I am rather pleased to announce my second publication of 2015, a tutorial paper on Bayesian modelling of perception.
Vincent, B. T. (2015) A tutorial on Bayesian models of perception, Journal of Mathematical Psychology. [manuscript pdf | supplementary pdf | code]
There are at least 3 different ways in which the Bayesian approach can be used: in data analysis models, with psychometric models, and, Bayesian Brain models which the paper focusses upon. I demonstrate how a simple perceptual task can be described as a probabilistic generative model (in 3 different ways), using the case study of the 2-alternative-forced-choice task.
This paper should be of interest to people who want to:
- understand the appeal of taking a Bayesian Brain approach,
- understand how probabilistic models can be constructed and used,
- learn how to implement probabilistic models in code. For this latter group of people who really want to get stuck into the code, I provide Supplementary Material as well as Matlab code on GitHub.
Along the way, a range of different topics in modelling and probability theory are introduced. While each individual topic may be covered in depth in the references I provide, I think I do a fair job of introducing a variety of topics including: Bayesian observer models, indirect perception, optimal vs. suboptimal Bayesian observers, generating simulated data, conducting parameter estimation of a simulated observer to experimental data, calculating model predictions, grid approximation, and MCMC sampling.
In a future post I will expand upon the 2 different evaluation methods I use in the paper: MCMC (using JAGS software) and grid approximation based on hand-coded evaluation of the model’s joint probability.
I hope this tutorial is useful for people, and I’m pleased to play a small part in promoting Bayesian models of perception. If you are super-interested in this stuff and the paper is not enough for you, then of course I am happy to deliver a seminar or workshop.