FLEXblock is an implementation of blockchain functionality for simulating descentralized federated learning experiments. It is intended to extend the FLEXible framework.
This repository includes both:
- An interface for creating your own blockchain architectures and working with them in a FLEX experiment.
- Some blockchain architectures to be used directly out of the box.
- flexBlock/pool: contains the interface for creating blockchain architectures and some implementations of it.
- flexBlock/blockchain: contains the interface for creating blockchains and block to be used with architectures and some implementations of it.
- notebooks: some explanatory notebooks showing how to implement a custom blockchain architecture and working with the library.
- tests: contains tests for the features present in this library.
- cifar-10.ipynb: Shows how to train a model in the cifar 10 dataset with a blockchain architecture.
- Proof of federated Learning MNIST.ipynb: Explains and shows how to use the Proof Of Federated Learning blockchain architecture for training a simple model in the popular MNIST dataset.
- Custom blockchain.ipynb: Shows how to create your own blockchain architecture and how to later use it in any kind of FLEX experiment.
The library implements an interface for creating your own blockchain architectures for FLEX experiments. Still, we offer some architectures out of the box.
| Architecture | Description | Citation |
|---|---|---|
| Proof of Work (PoW) | The most classical consensus mechanism used in blockchain where miners race into solving a hash puzzle. In this architecture the sets of miners(aggregators) and clients is disjoint. | Federated Learning Meets Blockchain in Edge Computing: Opportunities and Challenges |
| Proof of Stake (PoS) | This network uses a consensus mechanism where miners have some reputation/tokens called stake and they are picked randomly according to their stake. In this architecture the sets of clients and miners are the same. | Robust Blockchained Federated Learning with Model Validation and Proof-of-Stake Inspired Consensus |
| Proof of Federated Learning (PoFL) | A very popular consensus mechanism for blockchain enabled federated learning. Based on Proof of Useful Work miners race to get the model with the most accuracy on a given dataset. In this architecure the sets of clients and miners is disjoint. | Proof of federated learning: A novel energy-recycling consensus algorithm |
In order to install this repo locally:
pip install -e .
FLEX-block is available on the PyPi repository and can be easily installed using pip:
pip install flexBlock