A Practitioner's Guide to MXNet

Speaker:        Xingjian Shi
                Department of Computer Science and Engineering
                Hong Kong University of Science and Technology

Title:          "A Practitioner's Guide to MXNet"

Date:           Friday, 31 March 2017

Time:           4:30pm - 5:30pm

Venue:          Room 5510 (via lifts 25/26), HKUST

Abstract:

Over the past few years, we have witnessed the rise of a type of machine
learning method called deep learning (DL), which has achieved
state-of-the-art performances in many areas like natural language
processing, computer vision and speech recognition. Along with the fast
development of the field, the scale and complexity of DL models are
becoming increasingly large. It is thus essential to use DL libraries to
help understand and implement the existing models or develop new models.
In this presentation, I will introduce a DL library called MXNet. Compared
to other libraries like TensorFlow, Theano, Caffe and Torch, which either
only supports declarative programming or only supports imperative
programming, MXNet supports both programming paradigms and is more
flexible for DL practitioners. I will first give an overview of the
low-level and high-level APIs in MXNet and then go through three
programming examples: 1) Convolutional Neural Network, 2) Recurrent Neural
Network and 3) Parallel Training with Multiple GPUs. After that, I will
briefly introduce some advanced techniques including how to add new
operators and tricks to debug the program.


*******************
Biography:

Xingjian Shi is currently a 3rd year PhD student in Department of Computer
Science and Engineering, Hong Kong University of Science and Technology
supervised by Prof. Dit-Yan Yeung. Before that, he received his B.E.
degree from Shanghai Jiao Tong University in 2014. His research interests
include deep learning, spatiotemporal analysis and computer vision. He is
a member of DMLC and an MXNet committer.