Bayesian accounts of covert selective attention: a tutorial review

Screen Shot 2015-02-09 at 21.18.22

I am very happy to announce my new tutorial review paper.

Vincent, B. T. (2015) Bayesian accounts of covert selective attention: a tutorial review, Attention, Perception, & Psychophysics, 77(4), 1013-1032.

If you do not have an institutional subscription to Attention, Perception, & Psychophysics, Springer allow me to self-archive my author-accepted manuscript (legal). Get the preprints here: [manuscript pdf][supplementary pdf]. The final publication is available at Springer via http://dx.doi.org/10.3758/s13414-014-0830-0

This paper represents a first for me in a number of ways: it’s my first tutorial review, the first chance I’ve had to really explain the reason why I find the Bayesian approach exciting, and it’s the first paper where I’ve made my code available on GitHub.

Part 1 establishes the rationale behind low-level covert attentional experiments, provides an accessible introduction to why the Bayesian approach is interesting, and shows how attentional tasks can be approached from a decision making perspective.

Part 2 provides a tutorial overview of how we can model observers completing an attentional experiment from a decision making perspective. I am fairly happy with this as I present (as far as I can tell) the first graphical models of both the yes/no task, and alternative-forced-choice (AFC). As one reviewer pointed out, this model formulation really highlights the similarities between the two tasks. Readers interested in understanding the models in more depth can download the modelling code that implements Bayesian optimal observer modelling from this GitHub repository, https://github.com/drbenvincent/BayesCovertAttention.

Part 3 provides a thorough review and explanation of the literature of signal detection theory (SDT) and Bayesian models as applied to this category of simple, low-level covert attentional task.

I’m excited at having the chance to review the work done by others (and a little by myself), and to present some of my own wider ideas. 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.

Friendly questions about the code or bug reports are welcome through the GitHub repository (how to create a GitHub issue), or by email. Happy reading.

Leave a Reply