Biased Random-Key Genetic Algorithms: A Review (2312.00961v2)

Abstract: This paper is a comprehensive literature review of Biased Random-Key Genetic Algorithms (BRKGA). BRKGA is a metaheuristic that employs random-key-based chromosomes with biased, uniform, and elitist mating strategies in a genetic algorithm framework. The review encompasses over 150 papers with a wide range of applications, including classical combinatorial optimization problems, real-world industrial use cases, and non-orthodox applications such as neural network hyperparameter tuning in machine learning. Scheduling is by far the most prevalent application area in this review, followed by network design and location problems. The most frequent hybridization method employed is local search, and new features aim to increase population diversity. Overall, this survey provides a comprehensive overview of the BRKGA metaheuristic and its applications and highlights important areas for future research.

• The paper comprehensively reviews over 150 studies on BRKGA, emphasizing its unique random-key representation and biased mating strategies.
• It analyzes hybridization techniques, such as local search and multi-parent crossover, that enhance performance in scheduling, routing, and other applications.
• Challenges like premature convergence are addressed while outlining future research directions in multi-objective optimization and machine learning integration.

Biased Random-Key Genetic Algorithms: A Review

The paper authored by Londe et al. provides a comprehensive literature survey on Biased Random-Key Genetic Algorithms (BRKGA), a prominent population-based metaheuristic under the umbrella of genetic algorithms. The paper encompasses over 150 academic articles, scrutinizing the applicability and evolution of BRKGA in various domains, including complex combinatorial optimization problems, industrial applications, and non-orthodox scenarios such as neural network hyperparameter tuning.

Overview of BRKGA

BRKGA distinguishes itself by utilizing random-key-based chromosomes with a biased, uniform, and elitist mating strategy. The chromosomes are represented as vectors in a half-open unit hypercube, making the method agnostic to the problem structure and promoting code reuse. The decoder function plays a pivotal role by mapping individuals from the BRKGA space to the problem solution space, subsequently evaluating the solution's fitness.

Application Domains

The review identifies ten primary categories of application:

1. Scheduling: Nearly a third of BRKGA research targets scheduling problems. Hybridizing BRKGA with local search algorithms has been a trend, significantly improving solution quality and computational efficiency.
2. Network Configuration: Applications include weight setting problems in OSPF routing and network optimization. Indicator-based decoders and hybridization with local search are common.
3. Location Problems: These range from tollbooth problems to complex facility layout issues. Permutation-based decoders and hybrid methods, particularly local searches, are frequently employed.
4. Cutting and Packing: BRKGA is applied to container loading and bin packing problems, predominantly using permutation-based decoders without hybridization.
5. Vehicle Routing: Applications here highlight the importance of real-life-based instances and permutation-based decoders. BRKGA often outperforms other algorithms, though it occasionally lags behind Iterated Local Search (ILS) approaches.
6. Traveling Salesman Problem (TSP): BRKGA struggles with classic TSP due to its uniform crossover mechanism but performs better on generalized TSP versions.
7. Clustering: Indicator-based decoders are prevalent, and hybridization with clustering search and local search algorithms demonstrates significant performance gains.
8. Graph Problems: Applications cover clique problems and broadcast time issues. Permutation-based decoders are typical, with BRKGA consistently delivering high-quality solutions.
9. Parameter Optimization: BRKGA optimizes hyperparameters for machine learning models, often outperforming other metaheuristics.
10. Other Problems: Diverse problems such as non-linear equation solving, molecular docking, and COVID-19 mitigation strategies also benefit from BRKGA's versatility.

Hybridization Strategies

The survey emphasizes the varied hybridization strategies employed with BRKGA, which enhance its performance:

• Local Search Heuristics: Utilized both inside and outside the decoding phase, and often based on neighborhood moves such as item exchange, swap, and greedy insertion.
• Warm-Start: Initial populations are seeded with high-quality solutions generated from heuristics or mathematical models, significantly enhancing performance.
• Path Relinking (PR) and Differential Evolution (DE): These intensification methods are hybridized with BRKGA to explore new areas in the solution space.
• Multi-Objective Approaches: Adaptations for multi-objective problems often leverage concepts from effective algorithms like NSGA-II.

Novel Features and Challenges

Londe et al. discuss several new features introduced to the BRKGA framework to mitigate issues such as premature convergence:

• Island Model: Parallel evolution of populations with periodic exchange of elite individuals to enhance diversity.
• Reset and Shake Operators: These operators help in escaping local optima by reinitializing populations or disturbing elite solutions.
• Online Parameter Tuning: Adaptive mechanisms adjust parameters during evolution to balance exploration and exploitation.
• Multi-Parent Crossover and Implicit Path-Relinking (IPR): More sophisticated reproduction strategies to maintain diversity and explore promising regions in the solution space.

A notable challenge remains BRKGA's performance on problems with flat solution landscapes, where conventional methods outperform it unless specific adaptations are made, such as increased mutation rates.

Future Research Directions

The paper identifies several avenues for further research:

1. Multi-Objective and Stochastic Optimization: Developing frameworks specific to these problem types can unveil new applications and theoretical insights.
2. Hybridizations with Machine Learning: Leveraging machine learning for tasks like algorithm selection, fitness evaluation, and parameter tuning.
3. Integration with Other Metaheuristics: Exploring hybrids with algorithms like GRASP, ILS, and PSO.
4. API Development: Enhancing existing APIs and creating new ones in popular programming languages to facilitate broader adoption and experimentation.

Conclusion

BRKGA has proven to be a robust and adaptable metaheuristic across various problem domains. Hybrid strategies and new features continue to enhance its capabilities, making it a valuable tool in both theoretical and practical optimization scenarios. The paper by Londe et al. effectively highlights the strides made in BRKGA research and sets the stage for future developments in this dynamic field.