Advances in federated learning for healthcare & hyped blockchain project releases an SDK

Blockchain in healthcare updates

Substra: a framework for privacy-preserving, traceable and collaborative Machine Learning

The same party that last week open sourced their federated learning technology released a paper on their work this week. For some context again, Substra is the open source federated learning technology hosted under the Substra Foundation, and Owkin is implementing a version for healthcare enterprises in MELLODDY. For a brief description of federated learning I suggest you read last week’s issue or check out this comic from Google.

Related this week: A blockchain-orchestrated Federated Learning architecture for healthcare consortia

(Disclosure: I’m an author on this paper and a member of ConsenSys Health)

This paper is a high level overview of a novel architecture for federated learning that is designed with healthcare enterprises in mind. Several unique challenges come with this design remit, and an integration of multiple privacy preserving technologies is used to address these challenges. That includes:

  1. Fine grained data access policies

  2. An implementation of AMD’s SEV as an alternative to a Secure Aggregator build on multi-party computation

  3. Peer-to-peer encryption leveraging Ethereum identities for the passing of updated weights from workers to the Secure Aggregator

  4. A privacy preserving audit trail of events in a training round

I’d encourage you to read the paper to learn more about each of these design choices. Comments and feedback are also appreciated.

Regardless of the particular implementation, I think federated learning will be an important technology. It decouples the training of an algorithm from needing to share your data. This has immediate applications to enterprises in use cases like MELLODDY’s, but also long term impacts as users increasingly control and “own” their data. The future will look less like Google getting my health records or FitBit data without my consent or knowledge, and more like opt-in distributed AI networks built and owned by their participants.

An interview with MultiPlan CIO and SVP Michael Kim on blockchain and the Synaptic Health Alliance’s plans

Some select quotes that I found interesting, with ellipses indicating where a quote has been cut out of a broader answer:

HealthLeaders: Why has blockchain been so slow to take off in healthcare?

Kim: It's a convergence of a number of different factors. Blockchain has been slow to adoption [due to] technology, business process governance, and culture within the healthcare industry…

HL: Tell me about the Synaptic Health Alliance

Kim: There are six founding companies. Over the next month, a flurry of new folks will join the Synaptic Health Alliance, and it will continue to expand. We all have philosophically agreed that we want to create a utility in the marketplace, and that it's a not-for-profit organization…

HL: What results have you experienced?

Kim: Approximately 16% of the calls that [MultiPlan] makes from our outbound call centers today result in a change to the provider directory. As we participate in the Synaptic Health Alliance, our productivity [in Texas] has increased [about] four times that number. About 60% of [those] calls result in changes and updates to our provider directory information…

HL: What next steps will Synaptic Health Alliance take?

Kim: The pilot will wrap up during the first quarter of 2020, so we're near the end of that journey. It has been successful, and I'm cautiously optimistic given the level of interest that we have and new participants coming [into the Alliance]. Ultimately, the more participants there are, the more likely this is going to be successful.

There are three parts to Synaptic's growth. First, we're adding new states to the pilot, and we have very aggressive rollout plans. Second, we're adding new members to participate in the Alliance. The third growth [initiative] is tied to additional use cases…

HL: What challenges have you experienced?

Kim: The biggest issue associated with leveraging blockchain to solve problems in the healthcare industry is governance and standardization…

Podcast: Blockchain for Healthcare, Starting a Conversation with David Metcalf, PhD (Director METIL@UCF)

Article highlighting the work of IEEE in blockchain technology and clinical trials

What I’m reading this weekend

JPMorgan, Amazon, and Berkshire’s healthcare venture unveils their first move

Some details:

Unlike the existing insurance plan for JPMorgan’s U.S. employees, the new Haven programs don’t require employees to pay deductibles. They offer perks like earning money each month by fulfilling certain wellness activities such as keeping blood pressure below a certain target, said one of the people. That money can be used to offset doctor visits or the cost of prescriptions. Such benefits and incentives aren’t unusual in the corporate world.

Under Haven plans for the JPMorgan employees, co-pays range from $15 to $110 for most services, according to one of the people. Some more expensive care, such as a hospitalization, comes with higher charges.

Much hyped next-gen blockchain Dfinity released an SDK and new programming language

As the majority of the blockchain space has toned down their rhetoric and accepted the current scaling limitations blockchains have Dfinity stands out for their claims. They bill themselves as nothing less than The Internet Computer and a competitor to AWS itself. Dfinity technically differentiates itself with an algorithmic governance system and threshold relays, among other things. They raised $61m in early 2018 to built this out.

This weekend Dfinity released their first meaningful update in awhile, launching an SDK with the ability to deploy one local node. That is far from a mainnet deployment (or even private usage by an enterprise), but it is a step in the right direction. Anecdotally the programming language they’ve created has had a mixed reception, with some praise and some confusion on why they made the design choices they did.

NHSX released a report on getting artificial intelligence in healthcare right

A good read on artificial intelligence and policy implications in general.

This week was the 11th anniversary of the Bitcoin whitepaper

A less well known fact, but surely one that isn’t a coincidence, is that October 31st is also the anniversary of Martin Luther’s 95 theses.

A nice summary of the many different approaches to scaling a blockchain being worked on

Market Completeness and Arbitrage Freeness in decentralized finance

a16z Bio Newsletter: The Evolution of Gene Therapy

After President Xi’s announcement last week…

If you found this newsletter valuable then you can click the button below to sign up for free.

If you’re an existing reader I would deeply appreciate it if you share this with people who would find it a valuable resource. Feel free to reply to this email with comments, questions, or feedback. I also host a blockchain and healthcare Telegram group that you can join.