An Algorithmic Approach to Parameter Selection in Machine Learning using Meta-Optimization Techniques

摘要

The process of identifying the optimal parameters for an optimization algorithm or a machine learning one is usually costly, involves the search of a large, possibly infinite, space of candidate parameter sets, and may not guarantee optimality. Various attempts have been made to automate this process. Our work attempts to explore this research area further by analyzing the behavior of a simple genetic algorithm when used to find the optimal parameter setting for an ID3 like learner operating on a selected dataset.