Analysis of logic satisfiability in energy based discrete Hopfield neural network

Logic Satisfiability in conventional Discrete Hopfield Neural Network (DHNN) has suffered major issues such as the high potential to be trapped in suboptimal solutions. This cause problem to the network because suboptimal final neuron state that is trapped in a local minima solution will be disregarded as a potential solution for any given optimization problem. Energy-Based DHNN has the advantage to move the suboptimal final neuron state into a potential global solution. This type of DHNN utilizes temperature to change the position of the solution until the network achieves a global minimum solution. In this paper, we proposed a comprehensive comparison between two conventional energy-based DHNN in doing 2 Satisfiability namely Mean Field Theory DHNN and Boltzmann DHNN. The proposed network will be compared with conventional DHNN using various performance metrics and parameter settings. Finally, results from the experiment suggest that with the right parameter setup, the performance of Boltzmann DHNN are approximately equal to the optimal Mean Field Theory DHNN.

Text 121241.pdf - Published Version Restricted to Repository staff only Download (677kB) Official URL or Download Paper: https://pubs.aip.org/aip/acp/article-lookup/doi/10...

Actions (login required)

View Item