A Closer Look at Parameter Contributions When Training Neural Language and Translation Models
Raúl Vázquez, Hande Celikkanat, Vinit Ravishankar, Mathias Creutz, Jörg Tiedemann
Abstract
We analyze the learning dynamics of neural language and translation models using Loss Change Allocation (LCA), an indicator that enables a fine-grained analysis of parameter updates when optimizing for the loss function. In other words, we can observe the contributions of different network components at training time. In this article, we systematically study masked language modeling, causal language modeling, and machine translation. We show that the choice of training objective leads to distinctive optimization procedures, even when performed on comparable Transformer architectures. We demonstrate how the various Transformer parameters are used during training, supporting that the feed-forward components of each layer are the main contributors to the optimization procedure. Finally, we find that the learning dynamics are not affected by data size and distribution but rather determined by the learning objective.
BibTeX
@inproceedings{vazquez-etal-2022-closer,
title = "A Closer Look at Parameter Contributions When Training Neural Language and Translation Models",
author = {V{\'a}zquez, Ra{\'u}l and
Celikkanat, Hande and
Ravishankar, Vinit and
Creutz, Mathias and
Tiedemann, J{\"o}rg},
editor = "Calzolari, Nicoletta and
Huang, Chu-Ren and
Kim, Hansaem and
Pustejovsky, James and
Wanner, Leo and
Choi, Key-Sun and
Ryu, Pum-Mo and
Chen, Hsin-Hsi and
Donatelli, Lucia and
Ji, Heng and
Kurohashi, Sadao and
Paggio, Patrizia and
Xue, Nianwen and
Kim, Seokhwan and
Hahm, Younggyun and
He, Zhong and
Lee, Tony Kyungil and
Santus, Enrico and
Bond, Francis and
Na, Seung-Hoon",
booktitle = "Proceedings of the 29th International Conference on Computational Linguistics",
month = oct,
year = "2022",
address = "Gyeongju, Republic of Korea",
publisher = "International Committee on Computational Linguistics",
url = "https://aclanthology.org/2022.coling-1.424/",
pages = "4788--4800"
}