Research Area: Science of LMs
Keywords: Transformers, Nonparametric VIB, Reinterpretation, Post-training regularisation, Out-of-domain generalisation
TL;DR: We propose a novel reinterpretation of pretrained transformers which allows for an information-theoretic regularisation that improves out-of-domain generalisation on NLP tasks without retraining.
Abstract: Pretrained transformers have demonstrated impressive abilities, but tend not to generalise well out-of-domain and are very expensive to fine-tune on new domain data. Nonparametric Variational Information Bottleneck (NVIB) has been proposed as a regulariser for training cross-attention in transformers, potentially addressing this domain overfitting problem. We extend the NVIB framework to replace all types of attention functions in transformers. We show that existing pretrained transformers can be reinterpreted as nonparametric variational models using an empirical prior distribution and identity initialisation with controllable hyperparameters. We then show that changing the initialisation introduces a novel, information-theoretic post-training regularisation in the attention mechanism, which improves out-of-domain generalisation on NLP tasks without any additional training. This success supports the hypothesis that the way pretrained transformer embeddings represent information is accurately characterised by nonparametric variational Bayesian models.
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the COLM Code of Ethics on https://colmweb.org/CoE.html
Author Guide: I certify that this submission complies with the submission instructions as described on https://colmweb.org/AuthorGuide.html
Submission Number: 715
Loading