From White to Black through Gray... and one step back to Dark Gray

Evolutionary algorithms are one of the branches of artificial intelligence. They are used to solve a number of theoretical and practical optimization problems in various fields. Most often, these are difficult tasks of large size and complexity. One of the ways to improve the effectiveness of the optimization process is to decompose the structure of the problem. Many modern optimizers, including evolutionary algorithms, utilize the inter-gene dependencies model. It is useful in solving problems from various domains: binary, discrete non-binary, permutation-based, continuous, and others. It has been successfully applied to solving single- and multi-objective problems. A properly designed optimizer can significantly improve the quality of the final solution. Moreover, the appropriate use of the knowledge about the nature of the problem can lead to proposing results of high quality and at a low computational cost.

The purpose of this tutorial is to introduce participants to the four types of optimization: White-box, Gray-box, Black-box, and Dark Gray-box, and to show their similarities and differences, and their advantages and disadvantages. Then, we will show what the variable interactions graph is and we will discuss the origin and nature of overlaps. We will also show how you can significantly speed up the computation process. Next, we will discuss statistical and empirical linkage learning techniques. The discussed issues will be supported by numerous examples and drawings.

The tutorial will end with a summary in which we will indicate the most promising directions for further research.

This tutorial is dedicated to researchers and practitioners on all levels of expertise – to those starting their work with optimization or wishing to know more.