Improving Generalization in Meta-learning through Organization and Augmentation

Speaker:        Dr. Huaxiu Yao
                Stanford University

Title:          "Improving Generalization in Meta-learning through
                Organization and Augmentation"

Date:           Thursday, 22 April 2021

Time:           10:00 am - 11:00 am

Zoom Link:
https://hkust.zoom.us/j/465698645?pwd=c2E4VTE3b2lEYnBXcyt4VXJITXRIdz09

Meeting ID:     465 698 645
Passcode:       20202021


Abstract:

Meta-learning empowers an artificial intelligence agent to imitate how 
human beings continuously and quickly learn a task even with small labeled 
data. It has achieved notable success in diverse applications, such as 
image classification, question answering systems, and health risk 
prediction. However, the generalization ability of current meta-learning 
methods is limited by task heterogeneity and memorization. In this talk, I 
will first introduce two general principles to improve the generalization 
ability in meta-learning: organization and augmentation. Then, I will 
present several concrete instantiations of using each principle. This 
includes algorithms to organize and adapt knowledge continuously, a simple 
method for sufficiently overcoming task memorization, and several 
real-world applications. The remaining challenges and promising future 
research directions will also be discussed.


************************************

Biography:

Huaxiu Yao is currently a Postdoctoral Scholar of Computer Science at 
Stanford University, working with Chelsea Finn. His current research goal 
is to enable machine learning algorithms to learn quickly and efficiently 
via knowledge transfer. He is also passionate about applying these methods 
for solving real-world problems (e.g., smart city, healthcare, 
E-commerce). He obtained his Ph.D. degree from Pennsylvania State 
University. He also spent time in Amazon A9, Salesforce Research, Alibaba 
DAMO Academy, Tencent AI Lab, and Didi AI Labs. His research results have 
been published in top venues such as ICML, ICLR, NeurIPS, KDD, AAAI, and 
WWW.