LYNX recently ran a two-week hackathon in collaboration with Algovera.
Check out our kick-off video here:
The aim of the hack was three-fold:
- Test the “tech”: Specifically, Ocean Protocol’s compute-to-data functionality with an open-source EEG dataset on decentralised data storage (IPFS with estuary.tech & web3.storage).
- Test the “human” element: Can a decentralised group of people unite around the shared aim of building specialist algorithms, and together create ML capable of pulling out insights from brain data (EEG)?
- Kickstart our community: All who submitted an entry at the end of the hack received a LYNX POAP, enabling them to enter our Discord and continue to co-create LYNX. We also had four expert speakers (see the two talks below) from across the data ethics and neurotech space to help the community learn about these topics and catalyse ideation across these key themes.
The hack had 27 people sign-up from all over the world. This resulted in eight different teams. By day four, it became apparent that Ocean’s compute-to-data was not able to run on the large dataset. After learning this, we pivoted to give the teams access to the full data stored on IPFS.
The winning team - Team A made up of Arshy (@Arshy), Dr Prakash (@drprk.eth), Jimmy (@jneuro), Ben (@Ben) - did an excellent job at creating a classifier ML algorithm able to differentiate between eyes-open vs eyes-closed brain activity with an accuracy between 80% and 89%! You can see their code repository here.
Winning Team Walkthrough
- IPFS successfully served the decentralised data storage needs of the hack and provided a good user experience for researchers.
- Decentralised teams were able to successfully co-create specialist AI to pull out insights into brain activity together.
- Ocean’s compute-to-data functionality worked for the smaller sample file and hackers were able to use it. However, it is currently limited to smaller files (<200MB). Feedback on this functionality was obtained and communicated to Ocean. We will get there!
- We were able to pivot successfully and provided the large training & validation dataset (~2.5GB) via IPFS.
- There are a lot of incredible people out there who are interested in co-creating bespoke AI to pull out insights from biometric data! Thank you to all those who took part and have joined our community.
- Those previously unaware of compute-to-data found it a fascinating concept:
- We have had fantastic feedback on the hack overall:
“I really liked the way the hackathon was structured, the talks, and the intro call. The reaction to comments and unforeseen events. Also that there were the iPython notebooks already was helpful.”
We are really looking forward to growing out the LYNX community and partner with exciting and innovative wearables to further explore and co-create what is possible with this privacy-preserving technology.