Analysis of Evolutionary Diversity Optimisation for the Maximum Matching Problem

Jonathan Gadea Harder,Aneta Neumann,Frank Neumann

2024-04-18

Abstract:This paper explores the enhancement of solution diversity in evolutionary algorithms (EAs) for the maximum matching problem, concentrating on complete bipartite graphs and paths. We adopt binary string encoding for matchings and use Hamming distance to measure diversity, aiming for its maximization. Our study centers on the $(\mu+1)$-EA and $2P-EA_D$, which are applied to optimize diversity. We provide a rigorous theoretical and empirical analysis of these algorithms.

Neural and Evolutionary Computing

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to enhance the diversity of solutions in the maximum matching problem through evolutionary algorithms (EAs), especially for complete bipartite graphs and paths. Specifically, the goals of the paper are:

1. **Maximize diversity**: Use the Hamming distance as an indicator to measure diversity, and optimize the solution set in the matching problem through evolutionary algorithms to make it as diverse as possible.
2. **Analyze algorithm performance**: Conduct theoretical and empirical analyses on two evolutionary algorithms - (\( \mu + 1 \)-EA) and 2P - EA D to evaluate their efficiency in achieving maximum diversity in different situations.
3. **Provide running - time complexity**: Provide upper - bound estimates for the running time of these algorithms on complete bipartite graphs and paths.

### Main research content

- **Encoding and diversity measurement**: Represent the match with a binary string and use the Hamming distance to measure the difference between solutions.
- **Algorithm design**:
- (\( \mu + 1 \)-EA): Generate new solutions through mutation operations and remove the individual with the lowest diversity while maintaining the population size.
- 2P - EA D: Divided into two stages. First, "undo" the matches in a random subset, and then rematch these vertices to generate new solutions.
- **Theoretical analysis**:
- For complete bipartite graphs, the expected running time of (\( \mu + 1 \)-EA) in the small - gap case is given as \( O(\mu^2 m^4 \log(m)) \), and in other cases as \( O(\mu^2 m^2 \log(m)) \).
- For paths, the upper bound of (\( \mu + 1 \)-EA) is given as \( O(\mu^3 m^3) \), and the upper bound of 2P - EA D is given as \( O(\mu^3 m^2) \).
- **Experimental verification**: Through experiments, the influence of parameters \( m \) and \( \mu \) on the running time of the algorithm is studied, the theoretical results are verified, and the possibility of further improvement is proposed.

### Research significance

This research not only provides theoretical insights on how to effectively maximize diversity in the maximum matching problem, but also shows the practical application potential of these methods through empirical analysis. This is of great significance for scenarios that require diverse solutions, such as decision - support systems, combinatorial optimization, etc.