Seminar: The Role of Structure Modeling and Multi-task Learning for Neural Information Extraction

Seminar: The Role of Structure Modeling and Multi-task Learning for Neural Information Extraction

Room 212, E3 building, VNU - UET, 144 Xuan Thuy, Cau Giay, Hanoi

2019-09-18

2:00 pm - 4:00 pm

Thien Huu Nguyen

 

Abstract

Deep learning has transformed the field of natural language processing (NLP), offering the state-of-the-art models for a wide range of NLP tasks with better robustness and portability. The success of deep learning mainly comes from its ability to compose rich word representations to capture the underlying meanings of the input text via a variety of flexible neural network architectures.

A natural question at the moment is what we can do to further improve the performance of the deep learning models for the NLP tasks. Which limitations do the current deep learning models for NLP have? In this talk, I will present some recent research advances in deep learning for the task of information extraction (IE) in NLP. In particular, I will highlight the importance of structure modeling as well as the benefits of multi-task learning for different IE problems. In the first part of the talk, I will demonstrate how structure modeling can help to improve the compositionality and generalization of deep learning for relation extraction. The second part of the talk will focus on different techniques to perform multi-task learning for event extraction based on representation matching and task similarity. Finally, in the end of the talk, I will show some potential research directions for IE and deep learning as well as the research problems that we are working on at VinAI Research.

About the speaker

Thien Huu Nguyen is a Research Scientist at VinAI Research. Beside the work at VinAI, Thien is an Assistant Professor in the Department of Computer and Information Science at the University of Oregon. He obtained his Ph.D. and M.S. degrees in Computer Science from New York University (working with Ralph Grishman and Kyunghyun Cho), and his B.S. degree in Computer Science from Hanoi University of Science and Technology. He was also a postdoc in the University of Montréal, working with Yoshua Bengio and people in the Montreal Institute for Learning Algorithms.

Upcoming Events