Simulation of period-doubling behavior in distributed learning automata
E. Billard, S. Lakshmivarahan · 1998
This study presents simulation results on decisionmaking agents which operate using delayed information.For example, the delays can be due to latencies in a distributed computer network or because the .agentsbroadcast state information only intermittently or periodically.The decision process is modeled using a learning automaton, a rewardpenalty feedback mechanism.The simulation results, with the appropriate parameters, closely approximate the predicted oscillatory behaviors.As the penalty parameter decreases, the oscillations show period-doubling behavior and, at extreme settings, show the onset of chaos as the minima visit almost every point.Preliminary computation of the Lyapunov exponent also suggests chaotic behavior.