Domain Private Transformers for Multi-Domain Dialog Systems

Anmol Kabra, Ethan Elenberg


Abstract
Large, general purpose language models have demonstrated impressive performance across many different conversational domains. While multi-domain language models achieve low overall perplexity, their outputs are not guaranteed to stay within the domain of a given input prompt. This paper proposes domain privacy as a novel way to quantify how likely a conditional language model will leak across domains. We also develop policy functions based on token-level domain classification, and propose an efficient fine-tuning method to improve the trained model’s domain privacy. Experiments on membership inference attacks show that our proposed method has comparable resiliency to methods adapted from recent literature on differentially private language models.
Anthology ID:
2023.findings-emnlp.402
Volume:
Findings of the Association for Computational Linguistics: EMNLP 2023
Month:
December
Year:
2023
Address:
Singapore
Editors:
Houda Bouamor, Juan Pino, Kalika Bali
Venue:
Findings
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
6049–6061
Language:
URL:
https://aclanthology.org/2023.findings-emnlp.402
DOI:
10.18653/v1/2023.findings-emnlp.402
Bibkey:
Cite (ACL):
Anmol Kabra and Ethan Elenberg. 2023. Domain Private Transformers for Multi-Domain Dialog Systems. In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 6049–6061, Singapore. Association for Computational Linguistics.
Cite (Informal):
Domain Private Transformers for Multi-Domain Dialog Systems (Kabra & Elenberg, Findings 2023)
Copy Citation:
PDF:
https://aclanthology.org/2023.findings-emnlp.402.pdf