Investigating the learning behaviour of in-context learning: a comparison with supervised learning

Xindi Wang, Yufei Wang, Can Xu, Xiubo Geng, Bowen Zhang, Chongyang Tao, Frank Rudzicz, Robert E. Mercer, Daxin Jiang*

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference proceeding contributionpeer-review

2 Citations (Scopus)
30 Downloads (Pure)

Abstract

Large language models (LLMs) have shown remarkable capacity for in-context learning (ICL), where learning a new task from just a few training examples is done without being explicitly pre-trained. However, despite the success of LLMs, there has been little understanding of how ICL learns the knowledge from the given prompts. In this paper, to make progress toward understanding the learning behaviour of ICL, we train the same LLMs with the same demonstration examples via ICL and supervised learning (SL), respectively, and investigate their performance under label perturbations (i.e., noisy labels and label imbalance) on a range of classification tasks. First, via extensive experiments, we find that gold labels have significant impacts on the downstream in-context performance, especially for large language models; however, imbalanced labels matter little to ICL across all model sizes. Second, when comparing with SL, we show empirically that ICL is less sensitive to label perturbations than SL, and ICL gradually attains comparable performance to SL as the model size increases.

Original languageEnglish
Title of host publicationECAI 2023
Subtitle of host publication26th European Conference on Artificial Intelligence September 30–October 4, 2023, Kraków, Poland
EditorsKobi Gal, Ann Nowé, Grzegorz J. Nalepa, Roy Fairstein, Roxana Rădulescu
Place of PublicationNetherlands
PublisherIOS Press
Pages2543-2551
Number of pages9
ISBN (Electronic)9781643684376
ISBN (Print)9781643684369
DOIs
Publication statusPublished - 2023
Event26th European Conference on Artificial Intelligence, ECAI 2023 - Krakow, Poland
Duration: 30 Sept 20234 Oct 2023

Publication series

NameFrontiers in Artificial Intelligence and Applications
PublisherIOS Press
Volume372
ISSN (Print)0922-6389
ISSN (Electronic)1879-8314

Conference

Conference26th European Conference on Artificial Intelligence, ECAI 2023
Country/TerritoryPoland
CityKrakow
Period30/09/234/10/23

Bibliographical note

Copyright the Author(s) 2023. Version archived for private and non-commercial use with the permission of the author/s and according to publisher conditions. For further rights please contact the publisher.

Fingerprint

Dive into the research topics of 'Investigating the learning behaviour of in-context learning: a comparison with supervised learning'. Together they form a unique fingerprint.

Cite this