91 — Whale swarm algorithm for function optimization

This paper reviews a nature-inspired metaheuristic algorithm for optimizing NP-hard challenges. This “crossover” method emulates the ways in which whales communicate via ultrasound when hunting; the swarming behavior can be put to use as an optimization function.

Particle Swarm Optimizations (PSO) are swarm-intelligence systems based on the flocking behavior of animals such as birds or — hey hey! — whales. In PSO systems, individual “particles” traverse a parameter space based on individual, local functions. (Perhaps interesting to compare with the Co-fields Paper I covered earlier.)

In the Whale Swarm Algorithm (WSA), higher-fitness particles (“whales”) communicate when they have found “food” (a local optimum) by calling to other whales. The intensity $\rho$ of these calls are experienced at distance $d$ as $\rho = \rho_0 \cdot e^{-\eta \cdot d}$, with attenuation coefficient $\eta$. This means that a whale will follow low-distortion signals (i.e. from nearer whales) more avidly than those of lower-confidence interpretations of high-distortion calls from remote whales.

Though WSA is not guaranteed to return the global maxima, it is competitively fast with other nonexhaustive algorithms. It also returns competitive resultant maxima with other metaheuristic approaches.

Though this approach seems a bit contrived in its biomimicry, it’s also very clever: the discrete and lossy signals sent to other whales ensures that a single whale and its maximum of choice does not overtake the rest of the search. And the ability to easily parallelize this search system, using asynchronous communication between whales, is a large advantage in networked systems.