NeuraCrypt: Hiding Private Health Data via Random Neural Networks for\n Public Training Article Swipe

MAKE IMPACT

Adam Yala , Homa Esfahanizadeh , Rafael G. L. D' Oliveira , Ken R. Duffy , Manya Ghobadi , Tommi Jaakkola , Vinod Vaikuntanathan , Regina Barzilay , Muriel Médard · YOU? · · 2021 · Open Access · · DOI: https://doi.org/10.48550/arxiv.2106.02484

Balancing the needs of data privacy and predictive utility is a central\nchallenge for machine learning in healthcare. In particular, privacy concerns\nhave led to a dearth of public datasets, complicated the construction of\nmulti-hospital cohorts and limited the utilization of external machine learning\nresources. To remedy this, new methods are required to enable data owners, such\nas hospitals, to share their datasets publicly, while preserving both patient\nprivacy and modeling utility. We propose NeuraCrypt, a private encoding scheme\nbased on random deep neural networks. NeuraCrypt encodes raw patient data using\na randomly constructed neural network known only to the data-owner, and\npublishes both the encoded data and associated labels publicly. From a\ntheoretical perspective, we demonstrate that sampling from a sufficiently rich\nfamily of encoding functions offers a well-defined and meaningful notion of\nprivacy against a computationally unbounded adversary with full knowledge of\nthe underlying data-distribution. We propose to approximate this family of\nencoding functions through random deep neural networks. Empirically, we\ndemonstrate the robustness of our encoding to a suite of adversarial attacks\nand show that NeuraCrypt achieves competitive accuracy to non-private baselines\non a variety of x-ray tasks. Moreover, we demonstrate that multiple hospitals,\nusing independent private encoders, can collaborate to train improved x-ray\nmodels. Finally, we release a challenge dataset to encourage the development of\nnew attacks on NeuraCrypt.\n

Type

preprint

Landing Page

http://arxiv.org/abs/2106.02484

PDF

https://arxiv.org/pdf/2106.02484

OA Status

green

Cited By

9

Related Works

10

OpenAlex ID

https://openalex.org/W4322614810

Raw OpenAlex JSON

OpenAlex ID

https://openalex.org/W4322614810

Canonical identifier for this work in OpenAlex

DOI

https://doi.org/10.48550/arxiv.2106.02484

Digital Object Identifier

Title

NeuraCrypt: Hiding Private Health Data via Random Neural Networks for\n Public Training

Work title

Type

preprint

OpenAlex work type

Publication year

2021

Year of publication

Publication date

2021-06-04

Full publication date if available

Authors

Adam Yala, Homa Esfahanizadeh, Rafael G. L. D' Oliveira, Ken R. Duffy, Manya Ghobadi, Tommi Jaakkola, Vinod Vaikuntanathan, Regina Barzilay, Muriel Médard

List of authors in order

Landing page

https://arxiv.org/abs/2106.02484

Publisher landing page

PDF URL

https://arxiv.org/pdf/2106.02484

Direct link to full text PDF

Open access

Yes

Whether a free full text is available

OA status

green

Open access status per OpenAlex

OA URL

https://arxiv.org/pdf/2106.02484

Direct OA link when available

Concepts

Computer science, Adversary, Adversarial system, Robustness (evolution), Encoding (memory), Raw data, Suite, Machine learning, Autoencoder, Artificial neural network, Deep learning, Artificial intelligence, Encoder, Deep neural networks, Variety (cybernetics), Data mining, Information privacy, Computer security, Programming language, History, Biochemistry, Gene, Chemistry, Archaeology, Operating system

Top concepts (fields/topics) attached by OpenAlex

Cited by

9

Total citation count in OpenAlex

Citations by year (recent)

2024: 2, 2023: 3, 2022: 3, 2021: 1

Per-year citation counts (last 5 years)

Related works (count)

10

Other works algorithmically related by OpenAlex

Full payload