Melis, Gábor;
(2024)
Towards Better Generative Models of Language.
Doctoral thesis (Ph.D), UCL (University College London).
Preview |
Text
paper.pdf - Accepted Version Download (2MB) | Preview |
Abstract
As language modelling has been progressing immensely towards genuinely useful applications driven by the increased availability of data and computational resources, our understanding and tools have not kept pace. On the one hand, it is usual for understanding to lag behind practice; on the other, often a steep price is to be paid for letting practice race too far ahead. A trained language model is the product of the interactions of data, optimization, the training objective, regularization, and the model itself, which not only provide rich veins for research individually but combine to create significant complexity, which hampers progress. This thesis pursues advances along these directions to solidify the footing and our understanding of the model and the process of its training in hopes of guiding research and applications towards better solutions.
| Type: | Thesis (Doctoral) |
|---|---|
| Qualification: | Ph.D |
| Title: | Towards Better Generative Models of Language |
| Open access status: | An open access version is available from UCL Discovery |
| Language: | English |
| Additional information: | Copyright © The Author 2024. Original content in this thesis is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) Licence (https://creativecommons.org/licenses/by-nc/4.0/). Any third-party copyright material present remains the property of its respective owner(s) and is licensed under its existing terms. Access may initially be restricted at the author’s request. |
| UCL classification: | UCL UCL > Provost and Vice Provost Offices > UCL BEAMS UCL > Provost and Vice Provost Offices > UCL BEAMS > Faculty of Engineering Science UCL > Provost and Vice Provost Offices > UCL BEAMS > Faculty of Engineering Science > Dept of Computer Science |
| URI: | https://discovery.ucl.ac.uk/id/eprint/10185373 |
Archive Staff Only
 |
View Item |
