Doc.ai, a company that wants to use cryptocurrency to pay people for sharing their medical data with data scientists making predictive AI models, has acquired Crestle.ai, a platform for quick AI deployment. Doc.ai plans to pair Crestle with training for people in health care so that more physicians and medical practitioners can make predictive AI based on submitted data.
Crestle’s platform was made for one-click deployment of deep learning packages with Jupyter Notebook and is currently used by 11,000 data scientists and AI practitioners.
Financial terms of the deal were not disclosed.
Doc.ai raised $10 million in the fall of 2017 through a token sale supported in part by Ethereum cofounder Anthony Di Iorio and has since launched an app and digital wallet for users to receive Ethereum-based tokens. Doc.ai is currently awaiting information on compliance from federal regulators and hopes to begin giving token rewards to users, COO Sam De Brouwer told VentureBeat.
The first Doc.ai data trial involved more than 1,700 participants and health care companies like Anthem to predict seasonal allergies.
Doc.ai, as De Brouwer told VentureBeat, offers an open invitation to patients, physicians, data scientists, and health care companies who want to “join a little revolution” by sharing data that advances the state of AI in health care and delivers benefits to both individual participants and the health care industry.
“It feels like AI will be one of the main languages for all those stakeholders to be able to connect,” she said. “There’s a huge demand for physicians and data scientists working with medical data. This is where we want to focus to make sure the connection happens and symmetry is being built between the different stakeholders in health care.”
Training health professionals to train AI models
As part of the Crestle news, Doc.ai announced it will launch a course specifically for medical professionals in January 2019. The course will led by Doc.ai advisor and Harvard University assistant professor of bioinformatics Chirag Patel, as well as faculty member Arjun Manrai.
Both Crestle and Doc.ai have strong connections to Fast.ai, a startup dedicated to making deep learning more accessible for programmers.
Fast.ai offers a 7-week-long deep learning for beginners course led by Jeremy Howard, who is also chief science officer at Doc.ai. The online education initiative is designed to democratize AI and train anyone with basic coding knowledge to deploy AI models and has been taken by hundreds of thousands of people from around the world. Other courses, like one on how to use the Fast.ai deep learning library with PyTorch 1.0, are also available.
Crestle was created specifically for Fast.ai course participants by former student Anurag Goel. Other spinoff projects to come out of Fast.ai courses include the Envision app to help blind people identify objects and the computer vision-powered Not Hot Dog app on HBO TV show Silicon Valley, which was nominated for an Emmy.
Following a rewrite of the platform to make it capable of scale, Howard believes Crestle will be the best option for Fast.ai course attendees.
“We have to make sure that all the scalability works as students use it, but at the moment it’s looking like the best option for the one-click Jupyter Notebook approach. And then the full Linux environment approach, we’re currently recommending Google Cloud, because Google Cloud’s pricing is just fantastically good and they have an environment that’s all set up specifically for Fast.ai students,” Howard said.
Google Cloud Platform incorporates Fast.ai libraries, datasets, and lessons and can host a Linux server solution for course participants. AWS SageMaker support for Fast.ai course material is expected to become available in the weeks ahead, Howard said.
Participants in the Fast.ai Live course this fall and Fast.ai MOOC in January 2019 will be able to use Crestle GPUs in the cloud for training AI models free of charge.
Crestle is currently hosted by Google Cloud. The previous version was hosted by AWS, a company spokesperson told VentureBeat.