Hi,

I think your question should be asked more accurate. Generally, all the machine learning algorithms which are used for different generic goals (i.e., classification, clustering, regression) are proposed in order to solve a kind of optimization problems named data fitting. Indeed, the various learning algorithms are introduced for finding the suitable parameters of the target models (e.g., a binary classifier) by employing the given data (experiences). In simple words, the heart of machine learning is an optimization. Besides data fitting, there are are various kind of optimization problem. Moreover, over the last decades, different approaches were introduced in optimization problems for finding the best or satisfying solutions. Today, heuristic search strategies such as GA are usually used rather than exhaustive search, due to the feasible region sets of some optimization problems are very large. According to the type of optimization problems, machine learning algorithms can be used in objective function of heuristics search strategies.

A lot of paper use genetic algorithm to solve different kind of problems. The main advantage because you can at least find one application of this algorithm in your problem.

Pros

-Time to obtain result vs conventional algorithm

- Independence of kind of problem

-Easy to used in parallel (cluster, gpu, cpu)

-Access to libraries in different programing language

cons

-Problems with local optimum

-Not guaranties to reach global optimum

-Need to set parameters according with the problem

thank you for your answer Mr. Alan Palacio-Bonill . My question was about using machine learning approach(both supervised learning and unsupervised learning) for optimization problems, which means that the of the performance of algorithm will improve over time as the amount of and the quality of data increases compared to conventional optimization algorithms as you mentioned genetic algorithms..... which they have fixed performance.

Hi,

I think your question should be asked more accurate. Generally, all the machine learning algorithms which are used for different generic goals (i.e., classification, clustering, regression) are proposed in order to solve a kind of optimization problems named data fitting. Indeed, the various learning algorithms are introduced for finding the suitable parameters of the target models (e.g., a binary classifier) by employing the given data (experiences). In simple words, the heart of machine learning is an optimization. Besides data fitting, there are are various kind of optimization problem. Moreover, over the last decades, different approaches were introduced in optimization problems for finding the best or satisfying solutions. Today, heuristic search strategies such as GA are usually used rather than exhaustive search, due to the feasible region sets of some optimization problems are very large. According to the type of optimization problems, machine learning algorithms can be used in objective function of heuristics search strategies.

Thanks Mr. Hamid Bostani , very helpful explanation.

Good explanations at all.

Please find our recent work on using machine learning for optimization problems. As it is mentioned, most of the machine learning models often need to solve massive nasty optimization problems consisting of millions of parameters.

Preprint From feature selection to continuous optimization