Self-Regulated Sample Diversity in Large Language Models

Mingyue Liu, Jonathan Frawley, Sarah Wyer, Hubert P. H. Shum, Sara L. Uckelman, Sue Black and Chris G. Willcocks
Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics (NAACL), 2024

Core A Conference H5-Index: 133# Core A Conference

Self-Regulated Sample Diversity in Large Language Models
‡ According to Core Ranking 2023"
# According to Google Scholar 2023"

Abstract

Sample diversity depends on the task; within mathematics, precision and determinism are paramount, while storytelling thrives on creativity and surprise. This paper presents a simple self-regulating approach where we adjust sample diversity inference parameters dynamically based on the input prompt - in contrast to existing methods that require expensive and inflexible setups, or maintain static values during inference. Capturing a broad spectrum of sample diversities can be formulated as a straightforward self-supervised inference task, which we find significantly improves the quality of responses generically without model retraining or fine-tuning. In particular, our method demonstrates significant improvement in all supercategories of the MMLU multitask benchmark (GPT-3.5: +4.4%, GPT-4: +1.5%), which captures a large variety of difficult tasks covering STEM, the humanities and social sciences.

Downloads

YouTube

Citations

BibTeX

@inproceedings{liu24self,
 author={Liu, Mingyue and Frawley, Jonathan and Wyer, Sarah and Shum, Hubert P. H. and Uckelman, Sara L. and Black, Sue and Willcocks, Chris G.},
 booktitle={Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics},
 series={NAACL '24},
 title={Self-Regulated Sample Diversity in Large Language Models},
 year={2024},
 location={Mexico City, Mexico},
}

RIS

TY  - CONF
AU  - Liu, Mingyue
AU  - Frawley, Jonathan
AU  - Wyer, Sarah
AU  - Shum, Hubert P. H.
AU  - Uckelman, Sara L.
AU  - Black, Sue
AU  - Willcocks, Chris G.
T2  - Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics
TI  - Self-Regulated Sample Diversity in Large Language Models
PY  - 2024
ER  - 

Plain Text

Mingyue Liu, Jonathan Frawley, Sarah Wyer, Hubert P. H. Shum, Sara L. Uckelman, Sue Black and Chris G. Willcocks, "Self-Regulated Sample Diversity in Large Language Models," in NAACL '24: Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics, Mexico City, Mexico, 2024.

Supporting Grants

Similar Research

 

 

Last updated on 4 June 2024
RSS Feed