Control of annealing in minimal free energy vector quantization
D.R. Dersch, Paul Tavan · 2002
We derive conservation laws for minimal free energy clustering and vector quantization that allow one to monitor and control a corresponding annealing process. The controlled stochastic annealing thus obtained yields optimal solutions to that optimization problem, as we illustrate by sample simulations. We prove for the resulting codebooks that the discretization density is proportional to the data density in the limit of dense discretization.>