Making the Shoe Fit: Architectures, Initializations, and Tuning for Learning with Privacy
By Nicolas Papernot, Steve Chien, Shuang Song, Abhradeep Thakurta, Ú. Erlingsson. 2019
65 citations recorded by the source corpus.
Subfields: Optimization & Training, Privacy & Security
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Optimization & Training
- Peter Kairouz
- Abhradeep Thakurta
- Florian Tramèr
- Dan Boneh
- Om Thakkar
- Zheng Xu
- Shuang Song
- Yangsibo Huang
In Privacy & Security
- Peter Kairouz
- Abhradeep Thakurta
- Om Thakkar
- Florian Tramèr
- Zheng Xu
- Dan Boneh
- Shuang Song
- Gautam Kamath
Citations in the corpus
- Practical and Private (Deep) Learning without Sampling or Shuffling (2021, methodology): “[56] Nicolas Papernot, Steve Chien, Shuang Song, Abhradeep Thakurta, and Ulfar Erlingsson.”
- Differentially Private Learning Needs Better Features (or Much More Data) (2020, methodology): “For example, on CIFAR-10 we exceed the accuracy reported by Papernot et al. (2020b) while simultaneously improving the provable DP-guarantee by 130×.”
- Information Leakage in Embedding Models (2020, background): “The noisy training techniques are challenging for models with large parameters on large amounts of data and sensitive to the hyper-parameters [53].”
- Evaluating Gradient Inversion Attacks and Defenses in Federated Learning (2021, methodology): “Recent works [Papernot et al., 2020, Tramèr and Boneh, 2021] suggest using better pre-training techniques and a large batch size (e.g. 4 , 096 ) to achieve a better accuracy for DPSGD training.”
- Differentially Private Fine-tuning of Language Models (2021, background): “Consequently, smaller and simpler private models achieve the better accuracy in several settings (Papernot et al., 2019; Tram ` er & Boneh, 2021).”
- Large Language Models Can Be Strong Differentially Private Learners (2021, background): “For instance, Papernot et al. (2019); Tramèr & Boneh (2020); Abadi et al. (2016) consider transferring features learned on public datasets to simplify the subsequent private learning task.”
- AN EMPIRICAL STUDY (2022, unknown): “For CIFAR2 we follow the convention of fine-tuning a pre-trained model on a ‘public’ dataset [38] - in this case, we take a ResNet-56 trained on CIFAR-1003 and use its penultimate layer as an ‘embedding’ of CIFAR-10, resulting in examples of size d = 64.”