Inspiration
Crowdfunding, like Patreon and GoFundMe has provided to the ability for the public to directly fund driven people to accomplish things that could have been otherwise impossible. And with Patreon, the crowdfunder can additionally provide rewards to donors. However, the security and privacy of donors is of utmost importance but the current major public crowd public platforms cannot guarantee this. In fact, in 2015, Patreon has in fact been hacked, leaking 15GB of donor information [1]. And in 2014, usernames, e-mail addresses, mailing addresses and phone numbers, were leaked from Kickstarter [2]. In addition, currently, if the crowdfunder is aiming to create a product there is no option for crowdfunders to sell shares of the ownership of the product itself in a digitally verifiable way. Finally, if the product is digital, such as ML models, how can donors verify that the developers do not maliciously take other people’s models and treat it as their own, all the while collecting donations? Or from the developer’s perspective, how can they definitively prove that the model was trained by them, as a first step in chasing down the plagiarisers.
What it does
OlympiFund is a blockchain based platform designed for the development and crowdfunding of ML models. On the front page, potential donors can view the various ML projects. These potential donors can own shares of the project as well as see how much remaining shares are available. By using the blockchain, the ownership of each share is secure and verifiable. The developers can register a hash of their model on the blockchain, which acts as a verifiable and dated ownership record ML model. They can prove at the least that by the listed date, they had a model which hashes to the listed hash on the blockchain. If the anyone finds a model that hashes to the listed hash but was created later, then the developer can suspect their model was stolen.
How we built it
The front-end was built with react while the storage of the projects was implemented with smart contracts on the Blockchain Near. ML model hashes were obtained by hashing a string of the model state dictionary.
Challenges we ran into
We encountered roadblock in trying to incorporate Jackal blockchain for the ML Model storage. Jackal kept interfering with the rest of the program that was written with React.
Accomplishments that we're proud of
Able to create a working prototype in time.
What we learned
Being able to pivot is important.
What's next for OlympiFund
- Implement share transfer
- Hash out the details of how to transform amount of shares owned into monetary value. Use a royalty model? Customize number of shares? How would it fit with the rest of the financial ecosystem?
- What sort of business model will work for this high cost slightly niche market?
- How do we incorporate and combine more MLOPs with Blockchain? Use blockchain to store the whole model?
- Cross-chain payment
[1]https://arstechnica.com/information-technology/2015/10/gigabytes-of-user-data-from-hack-of-patreon-donations-site-dumped-online/ [2] https://www.cnet.com/news/privacy/kickstarter-hacked-user-data-stolen/
Built With
- jackal
- near
- react
- typescript
Log in or sign up for Devpost to join the conversation.