Lambda-opt neural approaches to quadratic assignment problems

Area of Science:

• Computational intelligence
• Operations research
• Artificial neural networks

Background:

• Combinatorial optimization problems, such as Quadratic Assignment Problems (QAPs), are computationally challenging.
• Existing heuristic methods can get trapped in local minima, limiting solution quality.

Purpose of the Study:

• To introduce novel analog neural network approaches for solving combinatorial optimization problems.
• To specifically address the challenges posed by large-scale Quadratic Assignment Problems (QAPs).

Main Methods:

• Development of analog neural network algorithms inspired by lambda-opt heuristics.
• Implementation of a middle-range search strategy by adjusting multiple assignments (lambda elements) simultaneously.
• Application of these methods to large-scale QAP instances (N = 80-150).

Main Results:

• The proposed analog neural methods demonstrate performance comparable to current state-of-the-art algorithms for QAPs.
• For specific benchmark problems, the new methods yielded superior solutions compared to previous champion algorithms.
• The approach effectively navigates the solution space, mitigating the issue of shallow local minima.

Conclusions:

• Analog neural network heuristics offer a promising approach for tackling complex combinatorial optimization problems.
• The developed lambda-opt based methods provide an effective strategy for escaping local minima in QAPs.
• These findings suggest potential for improved performance on large-scale optimization tasks.