Applying nonlinear learning scheme on AntNet routing algorithm

Abstract

The paper deals with a conceptual modification on the learning phase of AntNet routing algorithm through nonlinear reinforcement. Since the learning structure of AntNet consists of colonies of learning automata, the proposed approach replaces the previously defined linear learning automata structure with nonlinear learning automata, which modifies the reinforcement process without imposing overhead into the system. In order to select the appropriate nonlinear functions, the convergence rates are mathematically analyzed and the functions with better rates are replaced at the core of the system's learning cycle. To have an appropriate comparison four non-linear AntNet algorithms are considered and simulated on NSFNET topology, which are compared with the standard AntNet. Simulation results show that the vital performance metrics (e.g. packet delay, throughput, and network awareness) are improved using some forms of nonlinear learning functions.