Academic

PhD Oral Presentation

Feature selection is a fundamental method in many scientific discoveries in recent decades ranging from statistical analysis to machine learning applications. Due to the accelerated development of techniques, high-dimensional data and deep neural networks have become increasingly important in real-world situations. My research focuses on two parts: 1) propose the feature selection method, a generalization of correlation search, in two high dimensional regression models – instrumental variable regression model and varying coefficient regression model; 2) present an interpreting convolutional neural network (CNN) method – A-FMI, which uses the idea of feature selection, in order to tackle the limitations of the “black-box” nature of CNN.