Adaptive learning by extremal dynamics and negative feedback - art. no. 031912

We describe a mechanism for biological learning and adaptation based on two simple principles: (i) Neuronal activity propagates only through the netwo rk's strongest synaptic connections (extremal dynamics), and (ii) the stren gths of active synapses are reduced if mistakes are made, otherwise no chan ges occur (negative feedback). The balancing of those two tendencies typica lly shapes a synaptic landscape with configurations which are barely stable , and therefore highly flexible. This allows for swift adaptation to new si tuations. Recollection of past successes is achieved by punishing synapses which have once participated in activity associated with successful outputs much less than neurons that have never been successful. Despite its simpli city, the model can readily learn to solve complicated nonlinear tasks, eve n in the presence of noise. In particular, the learning time for the benchm ark parity problem scales algebraically with the problem size N, with an ex ponent k similar to 1.4.