, 2007; Wunderlich et al., 2012b); there are also powerful Pavlovian effects (Guitart-Masip et al., 2012). These might arise via dopamine’s hegemony over prefrontal-striatal interactions, possibly through the medium of parts of the dopamine system that are separable from those involved in functions buy Fulvestrant such as signaling reward prediction errors. It is certainly a general notion that (L) neuromodulators can play an important role in regulating internally directed computations (Robbins and Arnsten, 2009; Cools et al., 2011), and working memory has been a particular focus for this.

Serotonin also influences the activity of prefrontal neurons in rather complicated ways (Puig and Gulledge, 2011), potentially enabling it to influence executive operations such as working memory. The relationship between this and other possible functions of 5-HT such as predictions about punishment, is not yet clear. It is known that serotonin in the

orbitofrontal cortex is important for rapid adaption of behavior in paradigms in which inhibition of (possibly learned) prepotent responses is required (Roberts, 2011); and this can also be considered to be part of the regulation of internally directed computation. We discuss further aspects of this below. Utility is a poster child for the way that neuromodulators solve the communication problems raised in the introduction. It also shows well the scope and force of neuromodulation, which is very deeply embedded in the very structure GS-7340 of decision making. It is perhaps the intricacy of the interacting systems of modulation that is most conspicuous, with many of the general lessons reflecting combinations of architectural and receptor specificity, and also the substantial interdependence among the various parts. The representation, updating and use of uncertainty, have become major foci of computational treatments of neural information processing (Dayan et al., 2000; Doya et al., 2007; Ma and Pouget, 2008; Deneve, 2008, Körding, 2007; Fiser et al., 2010), with Bayesian analyses dominating. At a coarse time scale, organisms suffer

from ignorance about their environments, both because of limited opportunities to observe it, and because it changes in partly unpredictable ways. At a finer timescale, organisms have Resminostat to take noisy and partial observations from multiple sensory systems to estimate their circumstance in the world. This in turn influences the evaluation (and thus the execution) of actions, as we have just discussed. All of these facets lead to uncertainty, which in turn places severe constraints on what computations are normatively appropriate. Strict Bayesians admit no qualitative distinction between different sorts of uncertainty. However, strict Bayesian computations are usually radically intractable, and heuristics and approximations are necessary.