Data Science and Analytics Thrust Seminar | Towards Efficient and Robust NLP Systems

9:00am - 10:00am
Zoom ID: 967 9607 8205, Passcode: dsat

Artificial Intelligence (AI) systems are playing an increasingly important role in our daily lives, and it will also have significant impact to our future. One important aspect of this AI-powered future is that, we will be able to interact with these AI systems comfortably and confidently. To work towards such a future, we need Natural Language Processing (NLP) systems that can understand human language efficiently and robustly. In this talk, we present some recent works in this direction, including i) efficient Question Answering (QA) with Adaptive Computation, ii) a retrieval-based QA system with generate Probably-Asked Questions (winner solution of Neurips2020 EfficientQA competition), and iii) a data generation method to create debiased datasets to improve robustness NLI models. We end the talk with a discussion of efficient methods to incorporate large scale knowledge into NLP models.

Event Format
Speakers / Performers:
Yuxiang WU
University College London

Yuxiang Wu is a final-year PhD student in Computer Science at University College London, supervised by Prof. Sebastian Riedel and Prof. Pontus Stenetorp. His research interest lies at the intersection of Natural Language Processing (NLP) and Deep Learning, and his current focus is on Question Answering, Data Generation, and Knowledgeable Pretrained Language Models. He published eleven papers in top-tier AI/NLP conferences and journals (ACL, EMNLP, AAAI, IJCAI, TACL, etc.), and his work received more than 1259 citations (Google Scholar). He also won AKBC2020 Best Paper Award, Neurips 2020 EfficientQA Competition Champion in two tracks, and is a regular PC member of ACL/EMNLP/NAACL/AAAI/EACL conferences and various workshops.

Recommended For
Faculty and staff
General public
PG students
UG students
Data Science and Analytics Thrust, HKUST(GZ)
Post an event
Campus organizations are invited to add their events to the calendar.