A Survey on Networking for Neural Networks

PhD Qualifying Examination


Title: "A Survey on Networking for Neural Networks"

by

Mr. Junxue ZHANG


Abstract:

Over the past years, deep learning has driven tremendous progresses in computer 
vision, natural language processing and automatically driving. With 
increasingly large training datasets, the training time also grows, which 
sometimes even takes weeks to finish. To solve this, deep learning training has 
to go distributed. However, the improvement brought by the distribution is 
bounded by the networking performance.

In seeking high training efficiency for neural networks, it is natural to 
optimize the underlaying networking. In this survey, we first give an overview 
of neural networks and its relationship with networking. We also introduce the 
opportunities brought by emerging networking technologies. Then we discuss pure 
networking optimization approaches including improving communication patterns 
or adopting in-networking computation. Last, we take one step further and 
discuss application centric networking optimization which fully utilizes the 
properties of neural networks.


Date:			Tuesday, 12 June 2018

Time:                  	4:00pm - 6:00pm

Venue:                  Room 5508
                         Lifts 25/26

Committee Members:	Dr. Kai Chen (Supervisor)
 			Dr. Wei Wang (Chairperson)
 			Dr. Brahim Bensaou
 			Dr. Ke Yi


**** ALL are Welcome ****