Screenshot, Demo 4.
This program shows an adaptive-MPF hierarchy playing “Rocks, Paper, Scissors”. Two units are connected by a reward correlator. 's sensor-motor interface includes observed gestures and motor actions. The latter is a probability mass function (PMF) over the 3 possible gestures. These PMFs can be visualised as RGB values (red is rock, green is paper and blue is scissors). After a little learning, SOM models in respond to specific gestures but are not specific about the hierarchy's own actions. The motor PMFs are flat (so they appear close to greyscale values). Panel (a1) shows the sensor values of SOM models in . (a2) shows the motor values of SOM models in . (a4) shows the FB PMF with the roulette-selected action outlined in red. (b1) displays , the activation of SOM models in . (d1) and (e1) show the FF input and output of the RC. (e4) and (d4) show RC FB input and output respectively. (f1) shows SOM models in and (g1) shows for . (c2),(c3) show input and output of the first-order predictor in and (h2),(h3) the same for .