(Meta)Heuristic search algorithms gather a lot of sampling information as the search process proceeds. Can this information be used by machine learning algorithms to guide the search? For instance, to determine which search strategy is more effective or how to balance exploration and exploitation.
Dear Antonio Bolufé-Röhler,
It's an interesting question. Hybridization machine learning techniques with meta-heuristic algorithms are an interesting approaches on the improvement performance of these algorithms. Recently, some researches adopt clustering algorithms for global optimizations. For more details please refer to the following references. They can provide more motivations for you.
Best Regards,
Soheila.
http://www.sciencedirect.com/science/article/pii/S1568494610000840
http://www.sciencedirect.com/science/article/pii/S1568494606000809
http://www.sciencedirect.com/science/article/pii/S1568494606000846
http://ieeexplore.ieee.org/document/4220690/
Dear friend
Yes, ML is used for heuristic searches. I have used in my works. It provides better generalization on the learning parameters and the error rates are also reduced a lot. its used for optimization and lots of literature are available online. Here are a few that will help you.
Accurate prediction of coronary artery disease using reliable diagnosis system
Accurate telemonitoring of Parkinson's disease diagnosis using robust inference system
Machine learning algorithms for the creation of clinical healthcare enterprise systems
Developing new machine learning ensembles for quality spine diagnosis
Cheers
We have used learning for dynamically change the values of GRASP alpha parameter during the search process, in order to have control over "exploration" vs "exploitation" balance. You can search for "reactive grasp" for references.
Also, you may use a "probability of finding a better solution" notion for managing the parameter evolution. This probability is estimated using the known (or unknown) distribution of solution evaluations generated at each iteration of the algorithm.
Check out this paper, it presents several approaches on how to use supervized and unsupervized learning to improve metaheuristics.
Conference Paper Using Datamining Techniques to Help Metaheuristics: A Short Survey
- Badges
- Science topic