Automated Question Answering With Deep Learning & Neural Attention
Automated question answering (QA) has been one of the most important tasks in artificial intelligence community for several decades. While creating traditional QA systems required significant human efforts, the advancement of deep learning enabled us to train QA systems with minimal human intervention at a large scale. In this talk, I will present our recent work that achieved the state-of-the-art results on several QA tasks, showing that a machine can learn to read a long text, reason over multiple facts, and engage in a conversation with a human. Along with the presentation of other recent works, my talk will also overview the trend in and the future direction of QA research.
Minjoon is a 4th year PhD student in computer science at the University of Washington. His research focuses on natural language understanding and question answering. He is currently advised by Hannaneh Hajishirzi and Ali Farhadi, and he was also formerly advised by Oren Etzioni who is now the CEO of the Allen Institute for Artificial Intelligence. His work on designing a system for answering SAT geometry questions was featured in New York Times, Washington Post, and Geekwire. His recent work on deep neural networks for question answering on Wikipedia articles (SQuAD) achieved the first place in the dataset’s leaderboard.