Random forests, or randomized decision trees, are a popular ensemble predictive model, which have a rich and successful history in machine learning in general and computer vision in particular. Deep networks, especially Convolutional Neural Networks (CNNs), have become dominant learning models in recent years, due to their end-to-end manner of learning good feature representations combined with good predictive ability. However, combining these two methods, i.e., Random forests and CNNs, is an open research topic that has received less attention in the literature. A main reason is that decision trees, unlike deep networks, are non-differentiable. In this talk, I will introduce my recent work on integrating RFs with CNNs (Deep Random Forests) to address various machine learning problems, such as label distribution learning and nonlinear regression. I will show their applications to computer vision problems, such as facial age estimation. I will demonstrate how to learn the Deep Random Forests for different learning tasks by a unified optimization framework. The talk will start with a brief description of a few representative examples of my work.
Speaker Biography
Wei Shen received his B.S. and Ph.D. degree both in Electronics and Information Engineering from the Huazhong University of Science and Technology (HUST), Wuhan, China, in 2007 and in 2012. From April 2011 to November 2011, he worked in Microsoft Research Asia as an intern. In 2012, he joined School of Communication and Information Engineering, Shanghai University. From 2017, he became an Associate Professor. He is currently visiting Department of Computer Science, Johns Hopkins University. He has over 40 peer-reviewed publications in machine learning and computer vision related areas, including IEEE Trans. Image Processing, NIPS, ICML, ICCV, CVPR and ECCV.