Deep Learning with Induced Priors
TITLE:
Deep Learning with Induced Priors
DATE:
Friday, March 16th, 2018
TIME:
3:30 PM
LOCATION:
GMCS-314
SPEAKER:
Saining Xie, Doctoral Student, Computer Science, UCSD.
ABSTRACT:
Deep learning has reshaped the landscape of research and applications in
artificial intelligence. In contrast to traditional hand-crafted feature
engineering, and with the support of big-data and big-compute, the promise
and dominant paradigm of deep learning research is supervised, end-to-end
and automatic representation learning. However, to tackle many real world
problems, smart structural design decisions have to be made, oftentimes
through induced priors. Those priors can be hard-wired, or automatically
learned from the data or environment. In this talk I will introduce my
research in designing and utilizing deep learning structures for different
application scenarios in supervised learning and reinforcement learning.
In particular, I will talk about my recent work on induced priors and
structures for 3D point cloud recognition and transfer learning for
continuous control.
HOST:
Dr. Xiaobai Liu
DOWNLOAD: