Efficient emulation of neural networks on concurrent architectures for optimization problems

Specific optimization problems can be solved with high efficiency by neural network models, due to their intrinsic massive parallelism. The emulation of neural algorithms on concurrent architectures can preserve their computational power. The authors illustrate and discuss the implementation of Hopfield networks for the solution of the traveling salesman problem on 2-D meshes of transputers. The investigations are concentrated both on the computational efficiency and on the quality of solutions (global-local minima issue)