High Dimensional Feature Selection via a Slow Intelligence Approach

Feature selection in high dimensional space is the hot topic in contemporary machine learning area. In the past decade, a lot of effort has been devoted in developing various feature selection algorithms. However, each feature selection method has its own advantage for different datasets or applications. Therefore, one might face the difficulty of choosing the most suitable feature selection method in practice. Reference [1] proposed a new framework called Slow Intelligence System, which can continuously learn, search the appropriate method and propagate information according to the environment to improve the performance over time. In this paper, we adopt this slow intelligence idea in the application of high dimensional feature selection. We propose a new framework which could dynamically choose feature selection algorithm from a pool of methods according to the status quo. We apply our method to cancer datasets with the expression levels of thousands of genes as features. The experimental result shows that our method is superior to individual state-of-the-art feature selection methods. Keywords-Feature selection, machine learning, slow intelligence system