Imitation learning for structured prediction and automated fact checking
Andreas Vlachos
University of Sheffield
https://www.sheffield.ac.uk/dcs/people/academic/avlachos
15 November 2017
3:30pm - 4:30pm
EM G.44
Abstract
In the first part of this talk, I will introduce our work on imitation learning, a learning paradigm originally developed to learn robotic controllers from demonstrations by humans, e.g. autonomous helicopters from pilot's demonstrations. Recently, algorithms for structure prediction were proposed under this paradigm and have been applied successfully to a number of tasks such as dependency parsing, information extraction, coreference resolution and semantic parsing. Key advantages are the ability to handle large output search spaces and to learn with non-decomposable loss functions. In this talk I will give a detailed overview of imitation leaning, discuss its relation to other learning paradigms, describe some recent applications, including natural language generation, abstract meaning representation parsing and its use in training recurrent neural networks. In the second part of this talk, I will give an overview of our work on automated fact checking, and how it relates to the wider efforts on battling misinformation and rumour detection.
Bio
Andreas Vlachos is a lecturer at the University of Sheffield, working on the intersection of Natural Language Processing and Machine Learning. Current projects include natural language generation, automated fact-checking and imitation learning. I have also worked on semantic parsing, language modelling, information extraction, active learning, clustering and biomedical text mining. Previously, he was a postdoc at the Machine Reading group at UCL working with Sebastian Riedel and the BBC R&D team. Before that, he was a member of the NLIP group at the University of Cambridge working with Stephen Clark and a postdoc at the University of Wisconsin-Madison working with Mark Craven.
Hosts: Verena Rieser and Yannis Konstas