A Self Organized Classifier Based on Maximum Information Index and Its Development Using Vhdl

An entropy-based self organized learning algorithm is presented for classifier design. The method, which is based on ideas from information theory, maximizes the information index during feed forward learning in ANN. The neuron in the architecture divides the subspace based on the division of previous neurons which are connected to the current neuron by some probability distribution. Utilizing the entropy-based evaluator (EBE), the neuron further divides current subspace via optimal selection of the subspace control neuron, feeding features and fitting functions. Our method is simulated on standard benchmark examples and achieves performance improvement over many traditional classifier algorithm. The hardware development in behavioral level simulation using VHDL gains satisfactory results and provides significant reference to FPGA synthesis in structural level.