Abstract: 

Recent deep learning-based approaches have achieved remarkable success on a wide range of Natural Language Processing (NLP) tasks by learning from hundreds of thousands to millions of training data. Such models, however, hinge on the availability of large amounts of (annotated) data. Given the plethora of languages, tasks, and domains in the real world, where data is limited, such supervised learning frameworks break down. On the other hand, humans have a remarkable ability to learn quickly from a small number of examples, and extrapolate prior knowledge to new problems.
In this talk, I address the challenge of learning from limited data for a range of natural language understanding tasks and applications. I will present our work on few-shot learning approaches to NLP in both monolingual and cross-lingual settings and present findings in tasks such as word sense disambiguation, syntactic parsing and text classification. Finally, I will present recent research on approaches that can enable higher levels of data efficiency, and show how they can outperform much more computationally complex counterparts.

Bio: 

Helen Yannakoudakis is an Assistant Professor in Natural Language Processing (NLP) at the Department of Informatics, King's College London, and a Visiting Researcher at the Department of Computer Science & Technology, University of Cambridge. She is also a co-founder and Chief Scientific Officer at Kinhub (formerly Kami), translating research outcomes to deployable real-world applications in health and wellbeing. Her research focuses on machine learning for NLP, and specifically on transfer learning, few-shot learning, lifelong learning, multilingual NLP, and societal and health applications, such as language assessment, abusive language detection, misinformation, emotion and mental health detection. Helen is a Fellow of the Higher Education Academy, has received funding awards from both industry and academia, has won international competitions such as the NeurIPS 2020 Hateful Memes Challenge, and currently serves as an Area Chair for NeurIPS 2023.