Despite significant progress in AI, natural inteligence can still outperform AI in terms of learning efficiency and robustness, in a wide range of problems. There are also challenges to
adopt AI in domains at which explainabilty is a primary concern. Transfer learning is a primary research focus in the machine learning community to close the gap between natural inteligence and AI.
My research focus is transfer learning in general. More specifically, I am interested in learning in data-scarce regimes to improve the speed, efficiency, and robustness of machine learning
My current research interests include:
Lifelong learning is the ability of learning efficiently from a stream of data through benefiting from past experiences, and without forgetting the past learned knowledge. Compared
to many areas of AI, continual learning is still at its conception stage and there are many challenges that need to be addressed. Incorporating symbolic logic into continual learning is an area that I am very
Domain adaptation relaxes the need for data annotation in a "target domain" through transferring knowledge from a relevant "source domain", where annotated data is accessible.
Despite many existing works on domain adaptation, there are many application domains and subareas in domain adaptation that are not well-explored.
Low-Shot Leaning, including Zero-Shot and Few-Shot Learning
The goal in low-shot learning is acquiring the ability to recognize instances from novel classes using a small number of training instances from these classes.
Applications of transfer learning techniques
In application areas such as medicine and remote sensing, using data-driven deep learning is not feasible. Transfer learning can help to broaden the outreach of deep learning. We can also
use it to train explainable models to improve trust in AI.
UPenn Joseph D'16 and Rosaline Wolf Best PhD Dissertation Award
IJCAI Distinguished Student Paper Award
University of Waterloo Outstanding Graduate Achievement Award
Keston Research Award
ISI Exploratory Research Award
Please check the following if you are interested in joining my group:
I understand that many applicants would like to know more about the PhD program before the application deadline but
I apologize in advance that due to a large number of PhD application inquiries,
I cannot respond to all emails. If you are interested in working with me, please read about my research interests and mention my name on your PhD application.
I will check your profile after our deadline when the application pool is ready.
I may respond to a small number of emails if I feel that you have checked my publication well and consciously have narrowed down your research interests.
You can also email me after the application deadline to let me know your application is complete. Almost all decisions are done after the application deadline.
If you are already a master's student at USC and are interested in doing your thesis under my supervision,
please check my recent papers and contact me for further discussion.
Fruitful theses are normally the result of at least two semesters of continual effort.
USC Undergraduate Students:
Please contact me if you have already passed 2-3 courses on machine learning, data science, and artifcial inteligence.
Volunteer Remote Students:
Remote collaborations have become common recently and I have received inquiries for this purpose.
You can contact me for volunteer-based collaboration. I will try to define a research project for you but please note that remote collaboration requires
significant motivation on your end.
Fall 2021, Fall 2022: Applied Natural Language Processing (CSCI 544)
Fall 2020: Building Knolwedge Graphs (DSCI 558)