Mean Field Annealing for Pattern Classification using different response functions: A Comparative Approach.
Journal of ACS Advances in Computer Science · 2010
Mean Field Annealing (MFA) merges collective computation and annealingproperties of Hopfield Neural Networks (HNN) and Stochastic SimulatedAnnealing (SSA), respectively, to obtain a general algorithm for solvingcombinatorial optimization problems. Mean Field Annealing is adeterministic approximation, using mean field theory and stochasticsimulated annealing. Since MFA is deterministic in nature, this gives theadvantage of faster convergence to the equilibrium temperature, comparedto stochastic simulated annealing. The mathematics of MFA is shown toprovide a powerful and general tool for deriving optimization algorithms. Inthis paper, the MFA concepts are studied, the mathematics of MFA arederived, and different response functions are used to implement the MFAalgorithm. Experimental results are implemented using different networktopologies on a real classification problem known as Graph bipartitioningwhich was applied on Circuit Bi-partitioning. A comparative approachusing the different response functions is applied. Two annealing schedulesnamely: the Cauchy annealing schedule and the linear annealing scheduleare used and compared. The study and results are encouraging andpromising.