We studied the relative performance of stochastic heuristics in order to establish the relations between the fundamental elements of their mechanisms. The insights on their dynamics, abstracted from the implementation details, may contribute to the development of an efficient framework for design of new probabilistic methods. For that, we applied four general optimization heuristics with varying number of hyperparameters to traveling salesman problem. A problem-specific greedy approach (Nearest Neighbor) served as a reference for the results of: Monte Carlo, Simulated Annealing, Genetic Algorithm, and Particle Swarm Optimization. The more robust heuristics – with higher configuration potential, i.e. with more hyperparameters – outperformed the smart ones, being surpassed only by the method specifically designed for the task.
JavaScript jest wyłączony w Twojej przeglądarce internetowej. Włącz go, a następnie odśwież stronę, aby móc w pełni z niej korzystać.