Abstract
Neural Network(NN) is used popularly as discriminator in pattern recognition. In this paper, efficiency of NN activation function is validated the evidence in learning process and discrimination process. Sigmoid, Rectified and Linear are prepared as NN activation function. They functions use in singularity or combination. Learning method is Random Particle Swarm Optimization (RPSO) that is improved from standard PSO to can ferret in random order. In addition, digit data that is written hand from 0 to 9 are used as sample data. In conclusion, good results are gotten as discriminator in all patterns which used function in this paper. In the future, other functions that are better are found and difficult sample which is more digit sample are tried.
