Main Menu
— Event —

【Academic Seminar】Integer Programming in Machine Learning and Secure Artificial Intelligence - Dr. Baoyuan Wu

  • 2019.12.11
  • Event
Integer Programming in Machine Learning and Secure Artificial Intelligence

Topic: Integer Programming in Machine Learning and Secure Artificial Intelligence

Speaker: Dr. Baoyuan Wu, Tencent AI Lab

Time and Date: 14:00 - 15:00, Wednesday, December 11, 2019

Venue: Boardroom, Dao Yuan Building

 

摘要:

In this talk, I will present two research topics.

First, I will revisit the integer programming (IP) problem, which plays a fundamental role in many computer vision and machine learning applications. We propose a novel and versatile framework called Lp-box ADMM, based on the idea that the binary constraint is equivalent to the intersection of two continuous constraints. It reformulates the IP problem into an equivalent continuous optimization with only one non-convex constraint, which can be efficiently optimized by the standard ADMM algorithm. The Lp-box ADMM method is applicable to any integer programming tasks, such as image segmentation, clustering, model compression, hash code learning, etc. Second, I will give an overview of the literature of secure artificial intelligence, especially on adversarial examples. Based on the basic concepts of adversarial examples, I propose a general framework, of which most existing works of adversarial examples can be covered as special cases. Our several works under this framework will also be introduced, including black-box attack, physical attack, sparse attack, robust training, and adversarial detection, etc.

Above works were published in TPAMI and CVPR.

 

简介:

Baoyuan Wu is currently a Principal Research Scientist at Tencent AI Lab. He was Postdoc in IVUL lab at KAUST, working with Prof. Bernard Ghanem, from August 2014 to November 2016. He received the PhD degree from the National Laboratory of Pattern Recognition, Chinese Academy of Sciences (CASIA) in 2014, supervised by Prof. Baogang Hu. His research interests are machine learning and computer vision, including probabilistic graphical models, adversarial examples, multi-label learning and integer programming, etc. He has published 30+ top-tier (19 CCF-A) conference and journal papers, including TPAMI, IJCV, CVPR, ICCV, ECCV, AAAI, etc. Visit https://sites.google.com/site/baoyuanwu2015/ for more details.