Mısır, Mustafa2020-08-302020-08-302019Mısır, M. (2019, September). Active Matrix Completion for Algorithm Selection. In International Conference on Machine Learning, Optimization, and Data Science (pp. 321-334). Springer, Cham.9.78303E+120302-9743https://doi.org/10.1007/978-3-030-37599-7_27https://hdl.handle.net/20.500.12713/3135th International Conference on Machine Learning, Optimization, and Data Science, LOD 2019 -- 10 September 2019 through 13 September 2019 -- -- 236069Mısır , Mustafa (isu author)The present work accommodates active matrix completion to generate cheap and informative incomplete algorithm selection datasets. Algorithm selection is being used to detect the best possible algorithm(s) for a given problem ((formula presented) instance). Although its success has been shown in varying problem domains, the performance of an algorithm selection technique heavily depends on the quality of the existing dataset. One critical and likely to be the most expensive part of an algorithm selection dataset is its performance data. Performance data involves the performance of a group of algorithms on a set of instance of a particular problem. Thus, matrix completion [1] has been studied to be able to perform algorithm selection when the performance data is incomplete. The focus of this study is to come up with a strategy to generate/sample low-cost, incomplete performance data that can lead to effective completion results. For this purpose, a number of matrix completion methods are utilized in the form of active matrix completion. The empirical analysis carried out on a set of algorithm selection datasets revealed significant gains in terms of the computation time, required to produce the relevant performance data. © Springer Nature Switzerland AG 2019.eninfo:eu-repo/semantics/closedAccessConference Object11943 LNCS321334WOS:0006549425000272-s2.0-85078472934Q410.1007/978-3-030-37599-7_27Q3