@inproceedings{wu-etal-2022-adaptive,
title = "Adaptive Differential Privacy for Language Model Training",
author = "Wu, Xinwei and
Gong, Li and
Xiong, Deyi",
editor = "Lin, Bill Yuchen and
He, Chaoyang and
Xie, Chulin and
Mireshghallah, Fatemehsadat and
Mehrabi, Ninareh and
Li, Tian and
Soltanolkotabi, Mahdi and
Ren, Xiang",
booktitle = "Proceedings of the First Workshop on Federated Learning for Natural Language Processing (FL4NLP 2022)",
month = may,
year = "2022",
address = "Dublin, Ireland",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2022.fl4nlp-1.3/",
doi = "10.18653/v1/2022.fl4nlp-1.3",
pages = "21--26",
abstract = "Although differential privacy (DP) can protect language models from leaking privacy, its indiscriminative protection on all data points reduces its practical utility. Previous works improve DP training by discriminating privacy and non-privacy data. But these works rely on datasets with prior privacy information, which is not available in real-world scenarios. In this paper, we propose an Adaptive Differential Privacy (ADP) framework for language modeling without resorting to prior privacy information. We estimate the probability that a linguistic item contains privacy based on a language model. We further propose a new Adam algorithm that adjusts the degree of differential privacy noise injected to the language model according to the estimated privacy probabilities. Experiments demonstrate that our ADP improves differentially private language modeling to achieve good protection from canary attackers."
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="wu-etal-2022-adaptive">
<titleInfo>
<title>Adaptive Differential Privacy for Language Model Training</title>
</titleInfo>
<name type="personal">
<namePart type="given">Xinwei</namePart>
<namePart type="family">Wu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Li</namePart>
<namePart type="family">Gong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Deyi</namePart>
<namePart type="family">Xiong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2022-05</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the First Workshop on Federated Learning for Natural Language Processing (FL4NLP 2022)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Bill</namePart>
<namePart type="given">Yuchen</namePart>
<namePart type="family">Lin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Chaoyang</namePart>
<namePart type="family">He</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Chulin</namePart>
<namePart type="family">Xie</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Fatemehsadat</namePart>
<namePart type="family">Mireshghallah</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ninareh</namePart>
<namePart type="family">Mehrabi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tian</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mahdi</namePart>
<namePart type="family">Soltanolkotabi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xiang</namePart>
<namePart type="family">Ren</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Dublin, Ireland</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>Although differential privacy (DP) can protect language models from leaking privacy, its indiscriminative protection on all data points reduces its practical utility. Previous works improve DP training by discriminating privacy and non-privacy data. But these works rely on datasets with prior privacy information, which is not available in real-world scenarios. In this paper, we propose an Adaptive Differential Privacy (ADP) framework for language modeling without resorting to prior privacy information. We estimate the probability that a linguistic item contains privacy based on a language model. We further propose a new Adam algorithm that adjusts the degree of differential privacy noise injected to the language model according to the estimated privacy probabilities. Experiments demonstrate that our ADP improves differentially private language modeling to achieve good protection from canary attackers.</abstract>
<identifier type="citekey">wu-etal-2022-adaptive</identifier>
<identifier type="doi">10.18653/v1/2022.fl4nlp-1.3</identifier>
<location>
<url>https://aclanthology.org/2022.fl4nlp-1.3/</url>
</location>
<part>
<date>2022-05</date>
<extent unit="page">
<start>21</start>
<end>26</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Adaptive Differential Privacy for Language Model Training
%A Wu, Xinwei
%A Gong, Li
%A Xiong, Deyi
%Y Lin, Bill Yuchen
%Y He, Chaoyang
%Y Xie, Chulin
%Y Mireshghallah, Fatemehsadat
%Y Mehrabi, Ninareh
%Y Li, Tian
%Y Soltanolkotabi, Mahdi
%Y Ren, Xiang
%S Proceedings of the First Workshop on Federated Learning for Natural Language Processing (FL4NLP 2022)
%D 2022
%8 May
%I Association for Computational Linguistics
%C Dublin, Ireland
%F wu-etal-2022-adaptive
%X Although differential privacy (DP) can protect language models from leaking privacy, its indiscriminative protection on all data points reduces its practical utility. Previous works improve DP training by discriminating privacy and non-privacy data. But these works rely on datasets with prior privacy information, which is not available in real-world scenarios. In this paper, we propose an Adaptive Differential Privacy (ADP) framework for language modeling without resorting to prior privacy information. We estimate the probability that a linguistic item contains privacy based on a language model. We further propose a new Adam algorithm that adjusts the degree of differential privacy noise injected to the language model according to the estimated privacy probabilities. Experiments demonstrate that our ADP improves differentially private language modeling to achieve good protection from canary attackers.
%R 10.18653/v1/2022.fl4nlp-1.3
%U https://aclanthology.org/2022.fl4nlp-1.3/
%U https://doi.org/10.18653/v1/2022.fl4nlp-1.3
%P 21-26
Markdown (Informal)
[Adaptive Differential Privacy for Language Model Training](https://aclanthology.org/2022.fl4nlp-1.3/) (Wu et al., FL4NLP 2022)
ACL