Sudoku Solver by Q'tron Neural Networks

摘要

This paper presents a Sudoku solver based on the energy-driven neural-network (NN) model, called the Q'tron NN model. The rules to solve Sudoku are formulated as an energy function in the same form as a Q'tron NN's. The Q'tron NN for Sudoku can then be built simply by mapping. Equipping the NN with the proposed noise-injection mechanism, the Sudoku NN is ensured local-minima free. Besides solving Sudoku puzzles, the NN can also be used to generate Sudoku puzzles.