세미나안내

세미나안내

Why, How, What can a mathematician contribute in machine learning?

2019-11-07
  • 2,449
황형주 교수(POSTECH) / 2019.11.20

Biography

– Brown University (Ph. D): PDE, May 2002

POSTECH (MS) : Number Theory, Feb. 1997

POSTECH (BS) : Mathematics, Feb. 1995

-POSTECH (07 2006 – present) Assistant ~ Full Professor

PCAM(포스텍 수리응용센터) (01 2017 – present) Director

POSTECH Blockchain Technology Center(06 2018-Present), Vice Director

Brown University (01 2015 – 12 2015) Visiting Professor

Trinity College Dublin (09 2005 -07 2007) Permanent Lecturer (equiv. to Assistant Professor)

Max-Planck-Institute (07 2005 – 08 2005) Visiting Scholar

Duke University (08 2003 – 07 2005) Assistant Research Professor

Max-Planck-Institute (10 2002 – 07 2003) Postdoc

Abstract

Deep learning has shown remarkable achievements ever since hardware improvement enables heavy parallel computations through GPUs. Despite of its achievements in numerous fields such as image processing, natural language processing and etc, our theoretic understanding of its principles is far less studied compared to applications. In this seminar, we introduce some studies to figure out fundamentals of existing deep learning methodologies. First, we introduce the theoretic study of analytically showing convergence conditions of GANs. Second, we introduce the principled methodology to estimate reward functions in reinforcement learning. Lastly, we introduce the principled method to solve PDEs using neural networks.

목록