FELToken is connecting owners of the data with scientists to train their machine learning models while preserving the privacy of the data. Current federated learning solutions rely on a central server. We are building a product with the simplest user experience possible while being decentralized. This allows anonymous participation of data providers and prevents malicious activities. Data providers get rewards for sharing their data and resulting models can be further sold. Exchange of all models is encrypted so only authorized parties can use the model. We would like to further improve our tool by connecting with Ocean infrastructure.
Platform for a decentralized and more secure solution for federated learning while anonymizing data providers. Allowing data scientists to do machine learning on decentralized data without compromising its privacy.
- GitHub: Breta01 (Břetislav Hájek) · GitHub
- LinkedIn: https://www.linkedin.com/in/břetislav-hájek-75167111b
- GitHub: FilipMasar (Filip Masar) · GitHub
- LinkedIn: https://www.linkedin.com/in/filip-masar-776a0a174/
- Kernel KB4 fellow https://kernel.community/
- Co-founder at http://avtr.gg/
- Front-end developer at https://avast.com/
Using federated learning to train machine learning models across multiple datasets while preserving privacy using compute-to-data on Ocean marketplace.
The final product of FELToken is a platform for running machine learning across distributed data sets without compromising the privacy of the data. We already have our tool for running data provider clients and web applications for managing the projects. Now we would like to integrate it with the Ocean protocol. With Ocean integration, users will be able to use data already published on the Ocean marketplace using the compute-to-data. FELToken will act as an extra layer of functionalities over the ocean protocol. We would also like to add the possibility of further selling trained models through Ocean.
During development, we focus mainly on making the process of setting up the federated learning project as simple as possible. So that data scientists and interested parties can train their models without any extra knowledge about smart contracts and blockchain. Once the core components are working, there are many possible ways for further extension. The tool can also act as a platform for further federated learning research.
We want to focus on improving the web application and our python client code during this grant round. We have Ocean integration written in the form of python scripts. Over time, we want to transform these into our web application, making the usage of FELToken with Ocean simple for everyone.
- Update web application (dApp) - new design, better code design
- Clean the python code for Ocean integration
- Update Ocean integration for Ocean V4 (there are new features for setting environment variables for C2D)
- Write documentation for Ocean integration workflow
- Start working on demo with some other Ocean projects (DataUnion,…)
- Write blog post about our progress with Ocean
We have already finished the first release (MVP, video walkthrough: https://youtu.be/uoBl2yeO7hY). It demonstrates the core functionalities, and everyone can try it. We will continue working on more improvements and extensions. We have already finished the first version of Ocean integration:
Right now, we have the whole flow written in the form of Python scripts. We want to include this as part of our web application so that people can easily control the process through it.
Ocean marketplace integration will be a win-win situation for the Ocean and us. Our tool would add an extra analytics layer on top of the Ocean marketplace. We allow easier use of C2D and provide a way to compute valuable statistics or train machine learning on multiple datasets. And we will be able to attract more users who already use the Ocean for storing their data. With Ocean integration, we might bring even more users interested in machine learning to Ocean.