DERI Seminar with Priyanka Sen

Abstract: Question answering is the task of learning to predict answers to natural language questions. One way to train question answering models is using a knowledge graph, which is a structured representation of facts. Traditional approaches to knowledge graph-based question answering often rely on multiple models in a pipeline architecture. However this means that multiple models have to be deployed and maintained, and it’s difficult for individual models to learn from the successes and failures of the overall task. What if we could learn question answering end-to-end, using a single model? In this talk, we’ll look at the exciting work being done on end-to-end models using differentiable knowledge graphs to create scalable, interpretable, and competitive question answering models and share some of our group’s recent contributions in entity resolution and handling complex questions.

Bio: Priyanka Sen is a machine learning scientist at Amazon Alexa AI where she currently works in natural language processing research with a focus on question answering.