EMNLP Findings2022
Large-Scale Differentially Private BERT
Rohan Anil, Badih Ghazi, Vineet Gupta, Ravi Kumar, Pasin Manurangsi
Abstract
In this work, we study the large-scale pretraining of BERT-Large (Devlin et al., 2019) with differentially private SGD (DP-SGD). We show that combined with a careful implementation, scaling up the batch size to millions (i.e., mega-batches) improves the utility of the DP-SGD step for BERT; we also enhance the training efficiency by using an increasing batch size schedule. Our implementation builds on the recent work of Subramani et al. (2020), who demonstrated that the overhead of a DP-SGD step is minimized with effective use of JAX (Bradbury et al., 2018;Frostig et al., 2018) primitives in conjunction with the XLA compiler (XLA team and collaborators, 2017). Our implementation achieves a masked language model accuracy of 60.5% at a batch size of 2M, for = 5, which is a reasonable privacy setting. To put this number in perspective, non-private BERT models achieve an accuracy of ∼70%.
Ask Lune
Know this paper in minutes, not an afternoon.
Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.
47 top-tier papers have cited it since. Know what they found before your reviewers do.
Lune can draw on
the full paper · the top-tier papers that cite it · related work
Free to start. No credit card required.
Lineage
- Individual Privacy Accounting via a Rényi Filter cites Deep Learning with Differential Privacy
- Quantifying Memorization Across Neural Language Models cites The Secret Sharer: Evaluating and Testing Unintended Memorization in Neural Networks
- Quantifying Memorization Across Neural Language Models cites Deep Learning with Differential Privacy
- The Secret Sharer: Evaluating and Testing Unintended Memorization in Neural Networks cites Deep Learning with Differential Privacy
- Large Language Models Can Be Strong Differentially Private Learners cites The Secret Sharer: Evaluating and Testing Unintended Memorization in Neural Networks
- Large Language Models Can Be Strong Differentially Private Learners cites Differentially Private Fine-tuning of Language Models
- Differentially Private Fine-tuning of Language Models cites Large Language Models Can Be Strong Differentially Private Learners
- Large Language Models Can Be Strong Differentially Private Learners cites Deep Learning with Differential Privacy
- Differentially Private Fine-tuning of Language Models cites The Secret Sharer: Evaluating and Testing Unintended Memorization in Neural Networks
- Differentially Private Fine-tuning of Language Models cites Deep Learning with Differential Privacy
- Emergent and Predictable Memorization in Large Language Models cites Quantifying Memorization Across Neural Language Models
- Emergent and Predictable Memorization in Large Language Models cites The Secret Sharer: Evaluating and Testing Unintended Memorization in Neural Networks
- Emergent and Predictable Memorization in Large Language Models cites Deep Learning with Differential Privacy
Cited by top-tier papers47
- Large Language Models Can Be Strong Differentially Private LearnersXuechen Li, Florian Tramèr, Percy Liang, Tatsunori HashimotoICLR 2022 · 502 citations
- Differentially Private Fine-tuning of Language ModelsDa Yu, Saurabh Naik, Arturs Backurs, Sivakanth Gopi et al.ICLR 2022 · 494 citations
- Emergent and Predictable Memorization in Large Language ModelsStella Biderman, USVSN Sai Prashanth, Lintang Sutawika, Hailey Schoelkopf et al.NeurIPS 2023 · 205 citations
- Quantifying Memorization Across Neural Language ModelsNicholas Carlini, Daphne Ippolito, Matthew Jagielski, Katherine Lee et al.ICLR 2023 · 158 citations
- Individual Privacy Accounting via a Rényi FilterVitaly Feldman, Tijana ZrnicNeurIPS 2021 · 124 citations
Builds on13
- Deep Learning with Differential PrivacyMartín Abadi, Andy Chu, Ian J. Goodfellow, H. Brendan McMahan et al.CCS 2016 · 7,620 citations
- MLP-Mixer: An all-MLP Architecture for VisionIlya O. Tolstikhin, Neil Houlsby, Alexander Kolesnikov, Lucas Beyer et al.NeurIPS 2021 · 3,862 citations
- GShard: Scaling Giant Models with Conditional Computation and Automatic ShardingDmitry Lepikhin, HyoukJoong Lee, Yuanzhong Xu, Dehao Chen et al.ICLR 2021 · 1,954 citations
- The Secret Sharer: Evaluating and Testing Unintended Memorization in Neural NetworksNicholas Carlini, Chang Liu, Úlfar Erlingsson, Jernej Kos et al.USENIX Security 2019 · 1,386 citations
- Large Batch Optimization for Deep Learning: Training BERT in 76 minutesYang You, Jing Li, Sashank J. Reddi, Jonathan Hseu et al.ICLR 2020 · 1,170 citations
Related papers
- Enabling Fast Differentially Private SGD via Just-in-Time Compilation and VectorizationPranav Subramani, Nicholas Vadivelu, Gautam KamathNeurIPS 2021 · 96 citations
- Differentially Private Model CompressionFatemehsadat Mireshghallah, Arturs Backurs, Huseyin A. Inan, Lukas Wutschitz et al.NeurIPS 2022 · 18 citations
- Large Scale Private Learning via Low-rank ReparametrizationDa Yu, Huishuai Zhang, Wei Chen, Jian Yin et al.ICML 2021 · 122 citations
- Exploring the Limits of Differentially Private Deep Learning with Group-wise ClippingJiyan He, Xuechen Li, Da Yu, Huishuai Zhang et al.ICLR 2023 · 4 citations
- Training Text-to-Text Transformers with Privacy GuaranteesNatalia Ponomareva, Jasmijn Bastings, Sergei VassilvitskiiACL Findings 2022 · 9 citations
