Working memory for decisions


Dimitris Pinotsis


Using Nengo, a software package introduced at the workshop, simulate activity in PFC neurons engaging in a memory-to-decision transformation.


We model a well-known vibrotactile discrimination task (see e.g. Romo et al. 1999) where a monkey is asked to compare two mechanical vibrations (f1 and f2) separated by a time delay. We focus on both the memorization as well as the decision periods.


We built a discrimination circuit which consists of a comparator as well as a memory module. The former compares the input with what is stored in the memory and feeds a gated result back to the memory. The gating is mediated by a controller which manages the loading of the first and second frequencies as well as the delay period.

After the loading and memory phases, activity in the memory module will increase if the second frequency is f2<f1

or decrease if f2>f1:

Future Work

We will compare activity of simulated neurons with the activity recorded in the experiment of (Jun et al., 2010) focusing on the f2 decision period.

We will also extend our network formulation to deal with multidimensional inputs as opposed to a scalar input frequency considered here. Such an approach could predict neural activity in tasks where the encoded quantity can be represented in a higher dimensional space.