Two-and-a-half years ago, we set out on our journey to make decentralized, privacy preserving Artificial Intelligence as accessible as possible. At that time, while there were machine learning frameworks like PyTorch and Tensorflow, and there were cryptography libraries like SEAL and HELib, there were no open-source frameworks for general-purpose privacy preserving machine learning.
This is why our #1 priority has been the creation of free, commercially available, general purpose, open-source software (and education) for privacy preserving AI. The world already had privacy researchers (some even in machine learning). What has been most urgently needed is community, education, and tooling to make researching and deploying privacy preserving AI faster and easier to do. So, we became a community which, at its core, has been a development community.
30 months, 30 Repositories, 250+ Contributors, 3,000+ Pull Requests, 8,000+ Commits, 7,000+ Community Members, 12,000+ students, and $300,000+ in open-source grant funding later, we now live in a world with privacy-preserving machine learning libraries. As such, it is time to expand the scope of our community.
In addition to continuing to support open-source development, OpenMined will now also be a community for fostering novel research. We will have special emphasis on under-represented, social focused AI research: fairness, accountability, transparency, and (of course) privacy. However, as enabling secure access to private data is not merely a win for privacy, but also for the AI community in general, our research program is open to AI researchers of all interests and backgrounds.
The starting point for our research efforts is based on my personal observations of two under-served groups within our community.
First are researchers (especially graduate students) who have research experience, a strong research agenda, but only limited personal ability to tackle multiple research projects at once. It is my view that this group can often benefit from additional engineering support, especially engineers who are familiar with new tools for privacy preserving machine learning (such as PySyft).
Second are software engineers who would like to become researchers. Often it can be difficult to jump into a research community all on one's own. Papers are difficult to read and even more difficult to write without guidance for first publications from experienced researchers. Additionally, it can be difficult to gain entrance into strong academic programs (to then learn how to do research) without some existing research experience.
The starting point for OpenMined as a research community will be a matchmaking service between these two personas. We will seek to partner experienced developers who would like to work on a research project with experienced researchers who could accomplish greater amounts of research as a team and are willing to give guidance to collaborators along the way. We also expect all research teams to share co-authorship on resulting publications.
This matchmaking service will necessitate the formation of two new volunteer positions: Research Engineer and Research Scientist. Members of our research community will benefit from the following operations:
- Access to our matchmaking service - sourcing scientific and engineering collaborators within our community.
- Access to our core development teams for assistance, especially when working with our libraries.
- Access to a private "paper review co-operative" which will cross-review papers from amongst our community before major conference deadlines.
- The ability to list "OpenMined" as an official affiliation on academic papers.
- Finished research projects will be featured within OpenMined's content distribution network: our blog, twitter, youtube channel, social media, newsletter, and partner content distribution networks. This also includes content editing assistance, graphic design, and SEO optimization where applicable.
- Superior access to private datasets as a function of OpenMined's data partner network and privacy-preserving technology when relevant.
- Access to cloud compute credits from partner cloud providers when available and relevant.
- Access to research grants as they become available for specific projects.
However, while these are all wonderful things our community can provide, perhaps the greatest value is not any specific service the community offers, but the organic conversations the community necessarily creates along the way. In particular, I have found it an extremely common occurrence that we stumble upon very important research questions in our current activities. It is my great hope that in the near future, we can collaborate to help solve some of the most important challenges within AI and society, especially in fairness, accountability, transparency, and privacy.
Our first deadline which we will pursue as a research community is NeurIPS 2020. The paper submission deadline is on May 12th at 1pm PDT.
If you would like to join this community as either a Research Scientist or a Research Engineer, you may do so by filling out the corresponding application below:
I will be leading several research projects myself which have significant engineering components. I look forward to collaborating with you!