The Devil is in the Detail: Simple Tricks Improve Systematic Generalization of Transformers

Róbert Csordás, Kazuki Irie, Jürgen Schmidhuber

Research output: Chapter in Book/Report/Conference proceedingConference contribution

3 Scopus citations

Abstract

Recently, many datasets have been proposed to test the systematic generalization ability of neural networks. The companion baseline Transformers, typically trained with default hyper-parameters from standard tasks, are shown to fail dramatically. Here we demonstrate that by revisiting model configurations as basic as scaling of embeddings, early stopping, relative positional embedding, and Universal Transformer variants, we can drastically improve the performance of Transformers on systematic generalization. We report improvements on five popular datasets: SCAN, CFQ, PCFG, COGS, and Mathematics dataset. Our models improve accuracy from 50% to 85% on the PCFG productivity split, and from 35% to 81% on COGS. On SCAN, relative positional embedding largely mitigates the EOS decision problem (Newman et al., 2020), yielding 100% accuracy on the length split with a cutoff at 26. Importantly, performance differences between these models are typically invisible on the IID data split. This calls for proper generalization validation sets for developing neural networks that generalize systematically. We publicly release the code to reproduce our results.
Original languageEnglish (US)
Title of host publicationEMNLP 2021 - 2021 Conference on Empirical Methods in Natural Language Processing, Proceedings
PublisherAssociation for Computational Linguistics (ACL)
Pages619-634
Number of pages16
ISBN (Print)9781955917094
StatePublished - Jan 1 2021
Externally publishedYes

Fingerprint

Dive into the research topics of 'The Devil is in the Detail: Simple Tricks Improve Systematic Generalization of Transformers'. Together they form a unique fingerprint.

Cite this