Senge, Manuel ; Igamberdiev, Timour ; Habernal, Ivan (2022)
One size does not fit all: Investigating strategies for differentially-private learning across NLP tasks.
Konferenzveröffentlichung, Bibliographie
Kurzbeschreibung (Abstract)
Preserving privacy in contemporary NLP models allows us to work with sensitive data, but unfortunately comes at a price. We know that stricter privacy guarantees in differentially-private stochastic gradient descent (DP-SGD) generally degrade model performance. However, previous research on the efficiency of DP-SGD in NLP is inconclusive or even counter-intuitive. In this short paper, we provide an extensive analysis of different privacy preserving strategies on seven downstream datasets in five different ‘typical’ NLP tasks with varying complexity using modern neural models based on BERT and XtremeDistil architectures. We show that unlike standard non-private approaches to solving NLP tasks, where bigger is usually better, privacy-preserving strategies do not exhibit a winning pattern, and each task and privacy regime requires a special treatment to achieve adequate performance.
| Typ des Eintrags: | Konferenzveröffentlichung |
|---|---|
| Erschienen: | 2022 |
| Autor(en): | Senge, Manuel ; Igamberdiev, Timour ; Habernal, Ivan |
| Art des Eintrags: | Bibliographie |
| Titel: | One size does not fit all: Investigating strategies for differentially-private learning across NLP tasks |
| Sprache: | Englisch |
| Publikationsjahr: | 7 November 2022 |
| Ort: | Abu Dhabi, United Arab Emirates |
| Verlag: | Association for Computational Linguistics |
| Buchtitel: | Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing |
| URL / URN: | https://aclanthology.org/2022.emnlp-main.496/ |
| Kurzbeschreibung (Abstract): | Preserving privacy in contemporary NLP models allows us to work with sensitive data, but unfortunately comes at a price. We know that stricter privacy guarantees in differentially-private stochastic gradient descent (DP-SGD) generally degrade model performance. However, previous research on the efficiency of DP-SGD in NLP is inconclusive or even counter-intuitive. In this short paper, we provide an extensive analysis of different privacy preserving strategies on seven downstream datasets in five different ‘typical’ NLP tasks with varying complexity using modern neural models based on BERT and XtremeDistil architectures. We show that unlike standard non-private approaches to solving NLP tasks, where bigger is usually better, privacy-preserving strategies do not exhibit a winning pattern, and each task and privacy regime requires a special treatment to achieve adequate performance. |
| Fachbereich(e)/-gebiet(e): | 20 Fachbereich Informatik 20 Fachbereich Informatik > Ubiquitäre Wissensverarbeitung |
| Hinterlegungsdatum: | 30 Aug 2023 11:57 |
| Letzte Änderung: | 30 Aug 2023 12:08 |
| PPN: | 511165455 |
| Export: | |
| Suche nach Titel in: | TUfind oder in Google |
 |
Frage zum Eintrag |
Optionen (nur für Redakteure)
 |
Redaktionelle Details anzeigen |
Drucken
Impressum