Discussing the article: "Population optimization algorithms: Evolution Strategies, (μ,λ)-ES and (μ+λ)-ES"

[MetaQuotes]

Check out the new article: Population optimization algorithms: Evolution Strategies, (μ,λ)-ES and (μ+λ)-ES.

The article considers a group of optimization algorithms known as Evolution Strategies (ES). They are among the very first population algorithms to use evolutionary principles for finding optimal solutions. We will implement changes to the conventional ES variants and revise the test function and test stand methodology for the algorithms.

The new function is called "Hilly" (Fig. 2). Like "Forest" and "Megacity", it refers to complex test functions. For these three functions, the surface area lying above 50% of the maximum height is approximately the same and constitutes about 20% of the total area of the function.

The Hilly, Forest and Megacity functions provide complex and realistic optimization scenarios that can help evaluate the performance of algorithms under complex and varied conditions. By using these functions as a comprehensive test of optimization algorithms, we can gain more insight into their ability to find global optima and overcome local pitfalls.

In addition, changes have been made to the testing methodology. Now 10-fold testing is carried out instead of 5-fold (the number of repeated runs of the optimization process) to reduce random "spikes" in the results.

Author: Andrey Dik

[fxsaber]

That's a very big leap!

[Andrey Dik]

fxsaber #:
That's a pretty big leap!

Yes, an unexpected leap.

One could put it down to the fact that one of the functions was replaced, but no, it is still the best with Rastrigin. Just now, the overall complexity of the tests has increased and some algorithms have gone down, while those that were in the top remained there.