# A study of multi-parent crossover operators in a memetic algorithm

TL;DR: This work evaluates the performance of four multi-parent crossover operators (called MSX, Diagonal, U-Scan and OB-Scan) and provides evidences and insights as to why one particular multi- parent crossover operator leads to better computational results than another one.

Abstract: Using unconstrained binary quadratic programming problem as a case study, we investigate the role of multi-parent crossover operators within the memetic algorithm framework. We evaluate the performance of four multi-parent crossover operators (called MSX, Diagonal, U-Scan and OB-Scan) and provide evidences and insights as to why one particular multi-parent crossover operator leads to better computational results than another one. For this purpose, we employ several indicators like population entropy and average solution distance in the population.

## Summary (3 min read)

### 1 Introduction

- Memetic algorithms (MA) are known to be one of the highly effective metaheuristic approaches for solving a large number of constraint satisfaction and optimization problems [1].
- The literature reports a number of evolutionary and memetic algorithms with two-parent crossover operators for solving the UBQP problem ([3,4,5,6]).
- The authors are particularly interested in investigating the role of multi-parent crossover operators as well as a number of related important questions: why does one particular multi-parent crossover operator lead to better computational results than Corresponding author.

### 2.1 Main Scheme and Initial Population

- This study is based on the general memetic framework described in Algorithm 1 that alternates between a combination operator and a local improvement procedure.
- The combination operator (Section 2.4) is used to generate new offspring solutions while the local improvement procedure based on tabu search (Section 2.2) aims at optimizing each offspring solution.
- As soon as an offspring solution is improved by tabu search, the population is accordingly updated based on two criteria: the solution quality and the diversity of the population.

### 2.2 Tabu Search Procedure

- The authors employ a simple tabu search algorithm as their local search procedure.
- The authors tabu search procedure uses a neighborhood defined by the simple one-flip move, which consists of changing the value of a single variable xi to its complementary value 1− xi.
- The implementation of this neighborhood uses a fast incremental evaluation technique [9] to calculate the cost (move value) of transitioning to each neighboring solution.
- Tabu search incorporates a tabu list as a “recency-based” memory structure to assure that solutions visited within a certain span of iterations, called the tabu tenure, will not be revisited [10].
- The authors tabu search method stops when a given number α of moves are reached, called depth of tabu search.

### 2.3 Pool Updating

- In their memetic algorithm, after an offspring x0 is obtained by the crossover operator and improved by tabu search, the authors decide whether the improved offspring should be inserted into the population, replacing the existing worst solution.
- For this purpose, the authors define a quality-and-distance goodness score of the offspring x0 with respect to the population.
- Therefore, if the goodness score of the offspring solution is good enough, it will have high probability to replace the worst solution in the population.
- Interested readers are referred to [7] for more details.

### 2.4 Combination Operators

- The authors use four multi-parent crossover (or combination) operators to generate offspring solutions, including a “logic” multi-parent combination operator (MSX), a diagonal multi-parent crossover , a multi-parent uniform scanning crossover (U-Scan), a multi-parent occurrence based scanning crossover (OB-Scan).
- Thus, the variables with the first Avg largest Strength values receive assignment 1 and other variables receive assignment 0.
- The formal definition is given as follows.
- Thus each parent has the same probability to be the dominator of the value inherited by the offspring.

### 3.1 Instances and Experimental Protocol

- To evaluate the MSX, Diagonal, U-Scan and OB-Scan crossover operators, the authors carry out experiments on a set of 15 large random instances with 3000 to 5000 variables from the literature [13].
- The authors algorithm is programmed in C and compiled using GNU GCC on a PC running Windows XP with Pentium 2.66GHz CPU and 512MB RAM.
- Given the stochastic nature of the algorithm, problem instances are independently solved 10 times.
- The stop condition for a single run is respectively set to be 5, 10 and 20 minutes on their computer for instances with 3000, 4000 and 5000 variables, respectively.
- Note that when performing experiments on each crossover, the only difference consists in the crossover operator and other components of the algorithm are kept unchanged.

### 3.2 Computational Results

- Tables 1 and 2 report the best objective values (in parentheses number of hits over 10 runs) and the average objective values using the four crossover operators, respectively.
- The authors observe that MSX and U-Scan perform slightly better in terms of the best objective values since both two crossovers obtain the best values for 4 out of the 15 instances.
- OB-Scan seems to be the worst in terms of the best objective value and the success rate.
- Moreover, U-Scan and Diagonal perform quite well on 3 and 2 instances, respectively.
- In addition, the results also disclose that the performance of various crossover operators depend on the instances to be solved.

### 4 Analysis

- The above computational results show that for certain instances, some crossover operators perform better than other ones in terms of the solution quality.
- For this purpose, the authors introduce the following evaluation criteria to characterize the search capacity of different crossover operators: population entropy, average solution distance and average solution quality in the population.
- The authors also perform an experiment to show how different crossover operators and local search jointly influence the performance of the memetic algorithms.
- As an example, the experiments are presented on the large random instance p5000.5.
- From Tables 1 and 2, one observes that for this instance MSX performs the best, while U-Scan and OB-Scan are much worse than MSX and even Diagonal.

### 4.1 Evolution of Solution Quality

- The authors first study one of the most important characteristics for the four crossover operators, i.e., the solution gaps to the best known value evolving with the generation iterations, denoted by gb.
- Gb is defined as the average value of solution gaps between the best solution in the current population and the best known objective value over 10 independent runs.
- One observes that at the first generations, the four crossover operators have no clear difference in terms of this criterion.
- With the search progresses, MSX performs much better than other three ones.
- This observation coincides very well with the results reported in Tables 1 and 2, showing the advantage of the crossover operators of MSX and Diagonal, as well as the weakness of U-Scan and OB-Scan for this problem instance.

### 4.2 Population Entropy and Distance

- In their second experiment, the authors observe the two characteristics of the four multiparent crossover operators in terms of the population diversity: the population entropy vs. the number of generations; the average solution distance in the population vs. the number of generations.
- The authors see that the population diversity measured in terms of these two characteristics is better preserved during the evolution process for MSX and Diagonal than for U-Scan and OB-Scan, especially after the first 60 generations.
- Following the spirit of scatter search and path-relinking, an efficient solution combination operator is one that ensures not only high quality solutions but also a good diversity of the population.
- In other words, the diversification of the population induced by MSX and Diagonal allows the algorithm to benefit from a better exploration of the search space and prevents the population from stagnating in poor local optima.

### 4.3 Tradeoff between Intensification and Diversification

- The authors turn their attention to study another important aspect of the memetic algorithms, i.e., the tradeoff between local search and crossover operators.
- Under a limited computational resource, the depth of tabu search reflects the relative proportion of combination operators and tabu search in the algorithm.
- For each value, the authors perform 10 independent runs, each run being given 20 minutes CPU time.
- Figure 3 shows the average evolution of the best solution gaps during the search obtained with these three α values and four crossover operators.
- When it comes to the MSX crossover operator, one observes that MSX is not really sensitive to various α values, showing that MSX plays a real driving role for the search process.

### 5 Conclusions

- Understanding and explaining the performance of crossover operators within a memetic algorithm is an important topic.
- The authors presented an attempt to analyze the intrinsic characteristics of four crossover operators for the UBQP problem.
- To this end, the authors employed several evaluation indicators to characterize the search capability of a crossover operator.
- The experimental analysis allowed us to understand to some extent the relative advantages and weaknesses of the four studied crossover operators within the memetic framework.

Did you find this useful? Give us your feedback

...read more

##### Citations

38 citations

18 citations

### Cites result from "A study of multi-parent crossover o..."

...A study on multi-parent MA is found in (Wang et al., 2010) and they concluded that different combination got better results from others....

[...]

9 citations

6 citations

### Cites background or methods from "A study of multi-parent crossover o..."

...The offspring is constructed through extracting in a diagonal fashion correspondingly one section from each parent [29]....

[...]

...The choice can be uniform random choice or a random fitness proportional choice [27] [29]....

[...]

...This U-Scan breaks the traditional two parents approach and can extend the number of parents to an arbitrary number [29]....

[...]

...UNDX was implemented in the steady state genetic algorithm and it is used to solve three difficult large optimization problems [3][29]....

[...]

...In general, each parent votes and the values inherited will be the one privileged by the majority of the parents [29]....

[...]

4 citations

### Cites background from "A study of multi-parent crossover o..."

...Hence, the input for the crossover operator is two solutions and the output is two offspring’s....

[...]

##### References

42,654 citations

39,992 citations

12,322 citations

[...]

6,198 citations

1,457 citations