Probabilistic computational models of visual attention

14. Juli 2010

The problem of how to build computational models of visual attention and more generally, of cognition, will be discussed in its fundamental issues. In particular it will be shown how a probabilistic framework, and in particular a Bayesian approach, can provide a ground basis for such compelling task.

Remarkable progress in the mathematics and computer science of probability has led to a revolution in the scope of probabilistic models. Interestingly enough, the two viable solutions to approximate Bayesian inference, namely, the deterministic (e.g., Variational Bayes) and the Monte Carlo approximation techniques may also innocently lead to two rather different approaches for modelling visual attention, with respect to a basic, although apparently overlooked, controversial issue: the intrinsic randomness which makes a scanpath idiosyncratic to the observer and to the specific scene observed.

Examples either neglecting or accounting for such issue will be presented and discussed.