Sharing knowledge, not your data - Federated Learning

You, working from home

Imagine a world where all knowledge workers no longer commute to the office but instead work from home. Programmers, administrators, data scientists and more no longer have to move from their home to a central location every day, but instead stay put and simply transmit the outputs their employers need: their ideas, analyses, reports and other work products. Meanwhile the workers stay home and reduce time, cost, and risk of moving around and gathering.

We don't need to imagine it since we are living it. In January 2020, the current state of work-from-home would have seemed far-fetched at this scale, but now we are living it. Sure there were remote workers prior, but by necessity we have gotten past a whole lot of "this is how it was always done" inertia and found new ways of doing business for virtually the entire $4-6 TRILLION knowledge economy. And this isn't just temporary, big companies are paying upwards of $100 million to break their office lease and let their staff work remotely.

Permanently. This isn't going away. Certainly there will be adjustments and settling into what can be remote compared to what needs to be people moving around, but the dynamic has flipped - the cost and risk of in-person work now has to be justified, not the efficiency of remote work. "We have always done it this way," is no longer good enough.

Your data, working from home

Now imagine your health data working from where it sits: at the hospital, in the health specialist's office (or whatever server or cloud system they use), even sitting on your remote medical devices - without having to move. Vitals, physician diagnoses, referrals, insurance reimbursements, medical research all requiring no copying and transfer of your personal data. Simply transmit those key outputs without having to ever risk hacks, leaks and endless copies of your personal health information being shared all across the U.S. health system.

It may seem far-fetched and take a moment to get your head around, just like the current amount of telework and schooling would have back in January. Why are we making endless copies of some of our most personal (and highly regulated) data and sending it in bundles and complete datasets all over the world with little or no tracking? Why do we allow it to accumulate in huge centralized repositories that serve as financial incentive for bad actors to hack, steal, and exploit your data? It certainly isn't cheap or very effective, as our recent response to the pandemic has laid bare. And it doesn't keep our data safe as an never-ending news stream of hacks and data breaches shows (and those are only the ones we hear about).

"We have always done it this way," is unacceptable, because there is a better way.

Federated Learning

Federated machine learning or federated learning is the practice of training machine learning across distributed data in this way that has been rapidly expanding in health and other industries in the last several years. It has its roots in practices of federated registries and federated queries. These older formats for aligning distributed data sets are challenged by their analog nature and requirement of human compliance at different sites of data for quality control and assurance. Federated learning has replaced much of the noise and variability of this human element with the automated deployment of algorithms to directly query or learning from each pocket of local data and send back only those key derived elements to advance the learning. These derived data and related meta-data can provide a valuable machine learning environment with many of the data exposure concerns that arise from multiple copies and aggregation of valuable data. Fewer copies out there means less chance of a breach, and fewer large honey pots of valuable data lessens the incentive of malicious parties to hack and steal the data.

Blockchain-Orchestrated Federated Learning/Analytics

The next level of federated learning is the use of an underlying blockchain layer to help automatically maintain access control and a rapidly auditable ledger of every use or "touch" of any of the data involved, including potentially a record of everything that happens with the results derived from that data. How many drugs have been created utilizing your health data? How much money has been made on those drugs? That information has always technically been knowable, but the value proposition for doing it with legacy tech wasn't there. Now we are moving to a time where these questions are going to be much more easily knowable, and potentially rewardable.

Keeping your data in place, controlling access to it, tracking the results of that access, and having equity in the profits derived from those results are elements of the future that is now upon us.

A handful of efforts employing this blockchain and federated learning have begun worldwide for applications like the MELLODDY consortium for drug discovery. Beyond normal R&D challenges, the legacy issues of health data privacy and use of identifiable data for training and individual analytic queries has been limiting to how and were federated learning and analytics has been explored so far.

We have begun to unlock the full-potential of blockchain-orchestrated federated learning by adding layers of advance privacy/dignity preserving tech tools (e.g. zero knowledge proofs) as first described last year, "A blockchain-orchestrated federated learning architecture for health consortia." This idea begins to expand the applicability and promise of this model, using data without moving data or data at home. The roll-out of ConsenSys Health earlier this year has allowed for advances in development and piloting this approach. We had an in-depth discussion about this at our recent C19 Veterans Health Summit, "The art of the possible: How five emerging technologies will transform health." This was converted to a book chapter for publication in early 2021, and we will be presenting the latest on the topic at the IEEE AI Chain 2020 conference in Nov where we will present our accepted paper, "Blockchain-Orchestrated Machine Learning for Privacy Preserving Federated Learning in Electronic Health Data."

Worldwide HUG

At the ongoing Science Digital UNGA75 event paralleling the 75th United Nations General Assembly, we were able to show how this combination of technologies, in the right problem-solving and economic context, can go well beyond drug discovery and create the foundation for a whole new class of utility. The Health Utility Grid (HUG) was introduced to the world in the context of the UN's sustainable development goals in the presentation "Advancing Independence and Cooperation with the Health Utility Grid," and the subsequent "Health Utility Grid and Global SDGs" panel.

This new vision is, like the federated learning it employs, an inversion of the traditional legacy process. Not only does it allow your data to work safely from home, but also will allow patients world-wide to be in control of their own data while being compensated for its use. We are excited to be getting underway with this amazing effort and while 2020 in hindsight may have been a challenging year, the vision for the future is BRIGHT.