Binary optimization: Efficient increasing of global minimum basin of attraction
Iakov Karandashev, Boris Kryzhanovsky · Optical Memory and Neural Networks · 2010
The paper deals with the minimization of a quadratic functional in the configuration space of binary states. To increase the efficiency of the random-search algorithm, we offer changing the functional by raising the matrix it is based on to a power. We demonstrate that this brings about changes of the energy surface: deep minima displace slightly in the space and become still deeper and their attraction areas grow significantly. The experiment shows that use of the approach results in a considerable displacement of the spectrum of sought-for minima to the area of greater depths, while the probability to find the global minimum increases abruptly (by a factor of 10 3 –10 4 in the case of a two-dimensional Ising model).