How federated learning will drive a new economy of privacy-centric algorithms in healthcare.
How Federated Learning Ensures Patient Privacy
In medical school, one of the foundational tenets we are taught to uphold in the practice of medicine is a patient’s right to privacy. When it comes to healthcare data, privacy is not a nice to have or a want to have, it is a fundamental human right. We may accept consumer technology companies’ core business of tracking our movements and desires across cyberspace in the course of providing us with targeted advertising, but that’s not an acceptable approach to healthcare data, and never will be.
So how can healthcare leverage notoriously privacy-insensitive and data-hungry machine learning techniques like deep learning to transform our productivity at the same scale realized over the past decade in finance, industry, and consumer technology? Perhaps ironically, the solution to machine learning privacy in healthcare was first popularized by research scientists Brendan McMahan and Daniel Ramage in a 2017 Google blog post entitled, “Federated Learning: Collaborative Machine Learning without Centralized Training Data” Their approach was developed to address the mobile phone specific demands for, “smarter models, lower latency, and less power consumption, all while ensuring privacy” on Android, and was subsequently integrated into iOS by Apple. In the coming years, this approach to machine learning is expected to expand greatly beyond mobile phone use cases, since it is now supported by machine learning chip designers like NVIDIA for use in healthcare as well as other privacy centric use cases.
This article originally appeared on medium.com. Click here to read more.