FLock: World's 1st On-Chain Private Neural Processor

FLock = Federated Learning + Blockchain

Introduction

FLock is an open-source framework for building Fedrated Learning systems natively on Blockchain. The combination of Fedrated Learning and Blockchain solves real-life blockers for each other:

For Federated Learning : lack of participation incentives and risk of malicious clients can be solved by Blockchain technologies
For Blockchain : Highly consumed and wasted Proof of Work computation powers can be replaced by computations of Fedrated Learning models, aka, Proof of Machine Learning (PoML).

Principles

This project is built by a group of Oxford Alumni with strong industry and academic backgrounds in both Artificial Intelligence and Blockchain, which ensures the guidance of this project covers all different roles of participants in FLock:

For Developers :
- Build to earn
- "Own your data" vision
For Service Providers :
- Save bandwidth
- Maintain model performance
- Provably secure
For End Users :
- Passive income (run FLock to earn)
- Preserves privacy

Robustness

Our proposed v1 design is robust against attacks

Attack Method	Description	Countermeasures
Re-identification attack	Deriving encrypted identity information based on non-identity information in the data	No inter-node dataset transmission in FLock
Tracing attack	Deriving a person's presence without knowing their actual identification	No inter-node dataset transmission in FLock
Dataset reconstruction attack	Deriving a person’s attributes from a dataset without accessing the data	FLock processed gradients lacks enough information to support such attacks
Model-inversion attack	Observe model output under varied input circumstances to learn about unauthorized training data	FLock's gradient post-process make this attack computationally in tensive

Publications

Dong, N., Sun, J., Wang, Z., Zhang, S., & Zheng, S. (2022). FLock: Defending Malicious Behaviors in Federated Learning with Blockchain. doi:10.48550/ARXIV.2211.04344

More papers to come

Developments

We are currently undertaking most of developments in private mode, will add them here once they are released and made public:

FLock System Design paper: Check out this repo for details
FLock Healthcare App: Check out this repo for details
FLock Finance App: Check out this repo for details
FLock Model Distribution Serice: Check out this repo for details

Communications

Please star this project if you are interested, future communications and code/paper publications will be posted here.

We also welcome the community to join us in all possible ways!