GAMBIT: A Parameterless Model-Based Evolutionary Algorithm for Mixed-Integer Problems.

Area of Science:

• Optimization algorithms
• Computational intelligence
• Evolutionary computation

Background:

• Black-box optimization (BBO) presents challenges due to the lack of prior problem structure.
• Model-based Evolutionary Algorithms (EAs) excel at learning structure in discrete and continuous domains.
• Mixed-Integer (MI) optimization requires handling both discrete and continuous variables.

Purpose of the Study:

• To integrate discrete and continuous model-building mechanisms for the MI domain.
• To extend the Genetic Algorithm for Model-Based mixed-Integer opTimization (GAMBIT) with a parameterless scheme.
• To introduce a new mechanism for explicit exploitation of mixed dependencies in GAMBIT.

Main Methods:

• Integration of discrete and continuous model-building mechanisms within an EA framework.
• Development of a parameterless scheme for GAMBIT to enhance usability.
• Introduction of a novel mechanism for processing mixed variable dependencies.

Main Results:

• The parameterless GAMBIT demonstrates practical and efficient optimization capabilities.
• Explicit processing of mixed dependencies significantly improves optimization efficiency.
• GAMBIT successfully solves problems where variable dependencies hinder other algorithms.

Conclusions:

• The enhanced GAMBIT offers a robust and user-friendly approach to mixed-integer optimization.
• Exploiting mixed dependencies is crucial for advancing MI optimization techniques.
• GAMBIT shows competitive performance against state-of-the-art MI optimization algorithms.