报告摘要：
Various disciplines such as computer vision, natural language processing (NLP), speech or voice recognition, autonomous driving cars, etc., have recently been advanced revolutionarily, benefitting from the discovery of utilization of deep neural network with a purely data-driven manner. A deeply structured neural network that can accommodate significant non-linearity has been shown, in various fields, to outperform and generalize much better than conventional approaches that are usually conducive to an oversimplified phenomenological system. In this talk, I will be presenting general concepts of deep neural network such as learning objectives, supervised vs. unsupervised learning, different architectures of neural network such as feed-forward, convolutional, fully connected, as well as recurrent structures such as long short-term memory (LSTM), gated recurrent unit (GRU), hidden layer activation functions with issues such as vanishing gradient problems, dying rectified linear unit (ReLu) neuron, etc., normalization and regularization of the network, spatial information pooling,  discriminative vs generative approaches (generative adversarial network), etc. A few concrete examples of applications will also be shown to illustrate the training and learning procedures of a typical deep neural network with preliminary experimentation results reported.

报告人简介：Dr. Xing holds a Ph.D degree in Particle Physics, with his thesis studying b-quark physics at LHCb experiment at CERN. Thereafter Dr. Xing joined Stanford Linear Accelerator Center (SLAC) national laboratory as Engineering Physicist working on Free Electron Laser (FEL) X-ray physics analyses, focusing on using machine learning techniques to address issues that arise in experimentation data analysis. Dr. Xing then switched to the autonomous driving car industry as a Senior and then Staff Data Scientist at a leading China EV company, NIO. His main research focus is on utilizing deep learning techniques, Artificial Intelligence (AI) to facilitate some real-life problems within the self-driving cars domain such as car perception system, localization, as well as path planning systems.
第239期“论坛”主请人联系方式：
张黎明 17600853118  liming_zhang@tsinghua.edu.cn