A Goal-driven Approach for Combinatorial Optimization using Q’tron Neural Networks

T.-W. Yue and Z.Z. Lee (Taiwan)


Q’tron NN, Hopfield NN, knownenergysystem, solution qualifier, persisent noiseinjection mechanism, solution refining scheme, knapsack problem.


This paper gives an example (to solve the knapsack problem) to highlight the method to apply a Q’tron NN (neural network) model for combinatorial optimization. The Q’tron NN to solve the problem will be constructed as a known-energy system so that the NN will be dedicated to fulfill a particular goal-profit at any stage. The NN, as a result, will perform a goal-directed reasoning while the goal-profit is specified. Once the goal is fulfilled by the NN, we can supply another more profitable goal to the NN. In other words, the goal will be continuously refined, and, hence, the NN will report better and better solution progressively. Such a goal-refining is workable due to the Q’tron NN is intrinsically complete and local-minima free when it runs in full mode, i.e., noise injected.

Important Links:

Go Back