A principled approach to annealing in optimisation networks

Optimisation networks in their original form tended to produce poor, high cost solutions. To improve the quality of solutions, as measured relative to the cost of the optimal solution, it is necessary to guide the network state towards good solutions by the use of an annealing algorithm. The author shows that while hysteretic annealing can lead to improved solutions it may also lead to invalid solutions. Consequently, a new, principled approach to annealing is developed which retains the ability to guide the network state towards good solutions but can also guarantee that the network state remains valid