Neural network processing through energy minimization with learning ability to the multiconstraint zero-one knapsack problem

Defines a neural network model NNCO for the multiconstraint zero-one knapsack problem and presents a methodology to solve this problem approximately in near real-time by use of the characteristic that many neural networks can be shown to minimize an energy function defined for the networks. The authors overcome the difficulty that there is not guidance available as to what values the parameters ought to take by designing an algorithm, ADJUSTPAR, for automatically adjusting the values of the parameters used in the energy function. Moreover, they compare the present methodology with related work and demonstrate its advantages. They simulated the neural network model with several appropriate activation functions to solve approximately a set of 11 relatively large and difficult multiconstraint zero one knapsack optimization problems from the literature with well-known optimum solution. The result of the simulation demonstrates how the methodology can work well for the multiconstraint zero-one knapsack problem and can be easily extended to solve other combinatorial optimization problems