This research presents a performances analysis between different activation functions in solving non-systematic logical rule, specifically Random 2 Satisfiability (RAN2SAT) in Discrete Hopfield Neural Network. In this study, a new activation function called the Smish activation function will be integrated into the Discrete Hopfield Neural Network. The activation function plays a vital role in transforming the local field of the network into its final neuron state. In addition, the effectiveness of logic satisfiability in obtaining final neuron state depends on the type of activation function. The proposed new activation function will be compared with the conventional activation function, which is Hyperbolic Tangent activation function (HTAF) through computer simulations. The simulation of the different activation functions in doing logic satisfiability Discrete Hopfield Neural Network is done by DEV C++ version 15. Hence, the evaluation based on the different activation functions were made according to the error analysis, energy analysis, global minimum solution and total neuron variation. The output from the computer simulation shows that the logical rule of RAN2SAT with Smish activation function can retrieve the optimal solution. The finding of this research can give a benchmark for future research on non-systematic logical rule in doing logic satisfiability Discrete Hopfield Neural Network.
