Lu, Sheng ; Hendrik, Schuff ; Gurevych, Iryna (2024)
How are Prompts Different in Terms of Sensitivity?
2024 Conference of the North American Chapter of the Association for Computational Linguistics. Mexico City, Mexico (17-21.06.2024)
Konferenzveröffentlichung, Bibliographie
Kurzbeschreibung (Abstract)
In-context learning (ICL) has become one of the most popular learning paradigms. While there is a growing body of literature focusing on prompt engineering, there is a lack of systematic analysis comparing the effects of prompt techniques across different models and tasks. To address this, we present a comprehensive prompt analysis based on sensitivity. Our analysis reveals that sensitivity is an unsupervised proxy for model performance, as it exhibits a strong negative correlation with accuracy. We use gradient-based saliency scores to empirically demonstrate how different prompts affect the relevance of input tokens to the output, resulting in different levels of sensitivity. Furthermore, we introduce sensitivity-aware decoding which incorporates sensitivity estimation as a penalty term in the standard greedy decoding. We show that this approach is particularly helpful when information in the input is scarce. Our work provides a fresh perspective on the analysis of prompts, and contributes to a better understanding of the mechanism of ICL.
Typ des Eintrags: | Konferenzveröffentlichung |
---|---|
Erschienen: | 2024 |
Autor(en): | Lu, Sheng ; Hendrik, Schuff ; Gurevych, Iryna |
Art des Eintrags: | Bibliographie |
Titel: | How are Prompts Different in Terms of Sensitivity? |
Sprache: | Englisch |
Publikationsjahr: | Juni 2024 |
Ort: | Mexico City, Mexico |
Verlag: | Association for Computational Linguistics |
Buchtitel: | Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers) |
Veranstaltungstitel: | 2024 Conference of the North American Chapter of the Association for Computational Linguistics |
Veranstaltungsort: | Mexico City, Mexico |
Veranstaltungsdatum: | 17-21.06.2024 |
URL / URN: | https://aclanthology.org/2024.naacl-long.325 |
Kurzbeschreibung (Abstract): | In-context learning (ICL) has become one of the most popular learning paradigms. While there is a growing body of literature focusing on prompt engineering, there is a lack of systematic analysis comparing the effects of prompt techniques across different models and tasks. To address this, we present a comprehensive prompt analysis based on sensitivity. Our analysis reveals that sensitivity is an unsupervised proxy for model performance, as it exhibits a strong negative correlation with accuracy. We use gradient-based saliency scores to empirically demonstrate how different prompts affect the relevance of input tokens to the output, resulting in different levels of sensitivity. Furthermore, we introduce sensitivity-aware decoding which incorporates sensitivity estimation as a penalty term in the standard greedy decoding. We show that this approach is particularly helpful when information in the input is scarce. Our work provides a fresh perspective on the analysis of prompts, and contributes to a better understanding of the mechanism of ICL. |
Freie Schlagworte: | UKP_p_crisp_senpai |
Fachbereich(e)/-gebiet(e): | 20 Fachbereich Informatik 20 Fachbereich Informatik > Ubiquitäre Wissensverarbeitung |
Hinterlegungsdatum: | 24 Jun 2024 11:31 |
Letzte Änderung: | 02 Aug 2024 09:20 |
PPN: | 520293916 |
Export: | |
Suche nach Titel in: | TUfind oder in Google |
![]() |
Frage zum Eintrag |
Optionen (nur für Redakteure)
![]() |
Redaktionelle Details anzeigen |