English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Protecting large language models from privacy leakage is becomingincreasingly crucial with their wide adoption in real-world products. Yetapplying differential privacy (DP), a canonical notion with provable privacyguarantees for machine learning models, to those models remains challenging dueto the trade-off between model utility and privacy loss. Utilizing the factthat sensitive information in language data tends to be sparse, Shi et al.(2021) formalized a DP notion extension called Selective Differential Privacy(SDP) to protect only the sensitive tokens defined by a policy function.However, their algorithm only works for RNN-based models. In this paper, wedevelop a novel framework, Just Fine-tune Twice (JFT), that achieves SDP forstate-of-the-art large transformer-based models. Our method is easy toimplement: it first fine-tunes the model with redacted in-domain data, and thenfine-tunes it again with the original in-domain data using a private trainingmechanism. Furthermore, we study the scenario of imperfect implementation ofpolicy functions that misses sensitive tokens and develop systematic methods tohandle it. Experiments show that our method achieves strong utility compared toprevious baselines. We also analyze the SDP privacy guarantee empirically withthe canary insertion attack.

Just Fine-tune Twice: Selective Differential Privacy for Large Language Models