Studying generalisation using prompting for data augmentation
Guillem RamÃrez Santos (Ph.D. Student)
For the last five years, a new level of performance in NLP has been achieved by leveraging unsupervised training of large pretrained language models (PLM) and then fine-tuning on a downstream task. However, this is changing recently, as the community is shifting its attention towards prompting. The success of prompting shows that some PLMs have a good generalisation ability, meaning that they are able to understand and solve a task from limited examples or a brief description. I want to study a data augmentation method that is based on the generalisation ability; I believe this task will provide interesting insights on to what degree PLMs are able to generalise and extract commonalities between sentences. Studying this task could be beneficial to tackle higher level questions. What kinds of generalisations are PLMs able to capture? Is it possible to make PLMs become better at other types of generalisations? How can we best use this generalisation capability to improve few-shot learning?
| Primary Host: | Ivan Titov (University of Edinburgh & University of Amsterdam) |
| Exchange Host: | André Martins (University of Lisbon) |
| PhD Duration: | 12 September 2022 - 31 August 2026 |
| Exchange Duration: | 01 January 2024 - 01 July 2024 - Ongoing |
