Active Sensing as Bayes-Optimal Sequential Decision Making

Sensory inference under conditions of uncertainty is a major problem in both machine learning and computational neuroscience. An important but poorly understood aspect of sensory processing is the role of active sensing. Here, we present a Bayes-optimal inference and control framework for active sensing, C-DAC (Context-Dependent Active Controller). Unlike previously proposed algorithms that optimize abstract statistical objectives such as information maximization (Infomax) [Butko and Movellan, 2010] or one-step look-ahead accuracy [Najemnik and Geisler, 2005], our active sensing model directly minimizes a combination of behavioral costs, such as temporal delay, response error, and sensor repositioning cost. We simulate these algorithms on a simple visual search task to illustrate scenarios in which contextsensitivity is particularly beneficial and optimization with respect to generic statistical objectives particularly inadequate. Motivated by the geometric properties of the CDAC policy, we present both parametric and non-parametric approximations, which retain context-sensitivity while significantly reducing computational complexity. These approximations enable us to investigate a more complex search problem involving peripheral vision, and we notice that the performance advantage of C-DAC over generic statistical policies is even more evident in this scenario.