Mean Field Annealing for Pattern Classification using different response functions: A Comparative Approach. | ||||
| Journal of the ACS Advances in Computer Science | ||||
| Article 1, Volume 4, Issue 1, 2010, Page 1-31 PDF (720.09 K) | ||||
| Document Type: Original Article | ||||
| DOI: 10.21608/asc.2010.158214 | ||||
 View on SCiNiTO | ||||
| Abstract | ||||
| Mean Field Annealing (MFA) merges collective computation and annealing properties of Hopfield Neural Networks (HNN) and Stochastic Simulated Annealing (SSA), respectively, to obtain a general algorithm for solving combinatorial optimization problems. Mean Field Annealing is a deterministic approximation, using mean field theory and stochastic simulated annealing. Since MFA is deterministic in nature, this gives the advantage of faster convergence to the equilibrium temperature, compared to stochastic simulated annealing. The mathematics of MFA is shown to provide a powerful and general tool for deriving optimization algorithms. In this paper, the MFA concepts are studied, the mathematics of MFA are derived, and different response functions are used to implement the MFA algorithm. Experimental results are implemented using different network topologies on a real classification problem known as Graph bipartitioning which was applied on Circuit Bi-partitioning. A comparative approach using the different response functions is applied. Two annealing schedules namely: the Cauchy annealing schedule and the linear annealing schedule are used and compared. The study and results are encouraging and promising. | ||||
| Keywords | ||||
| Mean Field annealing; Hopfield Neural Network; Deterministic Annealing; Response Function; Ising Model; Annealing Schedule | ||||
|
Statistics Article View: 133 PDF Download: 174 |
||||
