Generalized Simulated Annealing

TL;DR

The paper introduces a generalized simulated annealing algorithm that efficiently finds global minima in complex, non-convex optimization problems, unifying and improving upon classical and fast annealing methods.

Contribution

It presents a new stochastic algorithm that generalizes existing simulated annealing techniques, offering potentially faster convergence for non-convex optimization tasks.

Findings

Recovers classical and fast simulated annealing as special cases

Can be more efficient than traditional methods

Applicable to high-dimensional non-convex problems

Abstract

We propose a new stochastic algorithm (generalized simulated annealing) for computationally finding the global minimum of a given (not necessarily convex) energy/cost function defined in a continuous D-dimensional space. This algorithm recovers, as particular cases, the so called classical ("Boltzmann machine") and fast ("Cauchy machine") simulated annealings, and can be quicker than both. Key-words: simulated annealing; nonconvex optimization; gradient descent; generalized statistical mechanics.