best way to balance data sharing and data privacy.
Data privacy is a very important topic these days.
Updated
On this page
Updated November 2026. This was first published on 17 November 2020. The idea is the same. Any number below is what the original piece said then. It is not a new count.
Key takeaways
- Data privacy is a very important topic these days.
- It gives rise to a very interesting challenge, since on one hand, Data Scientists are looking for high-quality models and accessible data.
- on the other hand, there is the need to keep data safe from both intentional and accidental leakage.
Data privacy is a very important topic these days. It gives rise to a very interesting challenge, since on one hand, Data Scientists are looking for high-quality models and accessible data. on the other hand, there is the need to keep data safe from both intentional and accidental leakage.
When doing Artificial Intelligence in the real world, you’ll find that most data sets come from large enterprises for two reasons: first, enterprises have a legal risk which prevents them from wanting to share their data set outside of their company. secondly, enterprises have a competitive advantage to hang on to large data sets collected from or about their customers.
This leads to one very challenging consequence. Though Data Scientists needs large amount of data to run Machine Learning models, they are often extremely constrained in terms of the amount of data they have access to solve their problems. This means that some of the most important things in society cannot be addressed with Machine Learning. Data Scientists don’t have access to the proper training data. using techniques for privacy preserving Artificial Intelligence such as Federate Learning and Differential Privacy, they can make a huge difference.
-What is Differential Privacy? #
Differential privacy is quite a new field. It recently started with statistical database queries around the begin of this Century and even more recently, has been applied to contexts such as Machine Learning.
The general goal of Differential Privacy is to ensure that different kinds of statistical analysis don’t compromise privacy. Privacy is preserved if, after the analysis, the analyst doesn’t know anything about the people in the data set.
But, in the context of statistics, and in all cases of Machine Learning it is insufficient. the goal is to learn something from a data set without learning specific things about people which might be sensitive or might harm them in some way.
A better definition of privacy is given by Cynthia Dwork, a pioneering figure in the field of differential privacy.
Differential Privacy describes a promise, made by a data holder, or curator, to a data subject (owner), and the promise is like this: **?You will not be affected adversely or otherwise, by allowing your data to be used in any study or analysis, no matter what other studies, datasets or information sources are available?.
-What is Federated Learning? #
Federated Learning is a type of remote execution wherein models are sent to remote data-holding machines for local training. This removes the need to store sensitive training data on a central server.
There are several use cases where Federated Learning is quite interesting. the most famous use case is in the context of mobile phones.
Your device downloads the current model, improves it by fitting from data on your phone, and then summarizes the changes as a small focused update. Only this update to the model is sent to the cloud, using encrypted communication, where it is immediately averaged with other user updates to improve the shared model. All the training data remains on your device, and no individual updates are stored in the cloud.
Federated Learning allows for smarter models, lower inactivity, and less power consumption, all while ensuring privacy. Moreover this approach has another immediate benefit: Also to providing an update to the shared model, the improved model on your phone can also be used immediately, powering experiences personalized by the way you use your phone.
This model become smart by looking at thousands and millions of other phones, without any of these people having spread the private information contained in their phone, and without you having spread any private information within your phone.
-What is the contribute of Federated Learning in Healthcare? #
Through Federated Learning, medical institutions are able to work better together and benefit from access to safe and cross-institutional data. Moreover smaller community and rural hospitals can access to expert-level AI algorithms.
Healthcare start-ups can bring innovations to market faster, thanks to a secure approach to learning from several algorithms. Meanwhile, research institutions are able to direct their work toward actual clinical needs, based on a wide variety of real-world data.
The University of Pennsylvania and Intel have built a federation of medical institutions to use Federated Learning to train Machine Learning models to identify boundaries of brain cancers. Brain cancers have multiple, sub-compartments that show up in different scans in different ways: active parts of the cancer that are really vascularized, necrotic parts of the cancer where tissue has died, and parts of cancers that are infiltrating healthy tissue. In this context the goal of Machine Learning is not to understand if the tumor exist or not, but to detect the boundaries of the cancer.
AI researchers from Nvidia and King’s College London have used Federated Learning to train a Neural Network for brain cancer segmentation. This method can allow data-sharing between hospitals and researchers while preserving patient privacy. The model uses a data set from the BraTS (Multimodal Brain Tumor Segmentation) Challenge of 285 patients with brain tumors.
Nvidia partnering with King’s College London and Owkin have created a Federated Learning platform for the National Health Service. The Owkin Connect platform running on Nvidia Clara enables algorithms to travel from one hospital to another, training on local data sets. It provides each hospital a blockchain-distributed ledger that captures and traces all data used for model training.
Researchers at MIT CSAIL, Harvard University Medical School, and Tsinghua University’s Academy of Arts and Design have developed one of the first approaches of Federated Learning to EMR (Electronic medical records) model training. This approach has an architecture that sources data from local hospitals, learns a model for each community, and aggregates the computed results on a server. This technique not only reduces data transmission costs between local hospitals and the model-hosting server, but it exposes dissimilarities among communities that otherwise might have escaped. The algorithm clustered a data set of 280.000 samples of patients into clinically meaningful communities with similar diagnosis and geographical locations, fitting one model for each community.
Enjoy with an online Federated Learning comic from Google AI.
Questions and answers #
-What is Federated Learning? #
Federated Learning is a type of remote execution wherein models are sent to remote data-holding machines for local training. This removes the need to store sensitive training data on a central server.
-What is the contribute of Federated Learning in Healthcare? #
Through Federated Learning, medical institutions are able to work better together and benefit from access to safe and cross-institutional data. Moreover smaller community and rural hospitals can access to expert-level AI algorithms.