# Federated Learning using Smart Contracts on Blockchains, based on Reward Driven Approach

**Authors:** Monik Raj Behera, Sudhir Upadhyay, Suresh Shetty  
**arXiv:** [2107.10243](https://arxiv.org/abs/2107.10243)  
**Date:** 2021-07-19  
**Summary page:** https://www.summarizepaper.com/en/paper/2107.10243v3/

## Key points

- The paper explores the use of federated machine learning in preserving data privacy while drawing insights from data
- One challenge in adopting federated learning is the lack of fair and transparent incentivization schemes for rewarding contributors
- The authors propose leveraging smart contracts on a blockchain network to define incentivization rules based on a scalar quantity called federated contribution
- Smart contracts provide transparent, immutable, and independently verifiable proofs by all participants in the network
- Using smart contracts on blockchains ensures fairness, transparency, and universality in incentivization schemes while preserving data privacy
- The proposed approach has potential applications in compliance and anti-money laundering cases where multiple sensitive datasets need reconciliation and fraud/anomaly detection models can enhance audits and investigations
- The authors emphasize that their approach leverages a unique structured ledger with potential privacy layers to guarantee security, accuracy, and correctness in distributing the model's parameters
- Current blockchains require privacy layers to protect communications between participants during validation

## Summary

The paper titled "Federated Learning using Smart Contracts on Blockchains, based on Reward Driven Approach" explores the use of federated machine learning in preserving data privacy while drawing insights from data. One of the challenges in adopting federated learning is the lack of fair and transparent incentivization schemes for rewarding contributors. To address this issue, the authors propose leveraging smart contracts on a blockchain network to define incentivization rules based on a novel scalar quantity called federated contribution. Smart contracts provide transparent, immutable, and independently verifiable proofs by all participants in the network. By utilizing the open and transparent nature of smart contracts, a reward-driven model can be established to revolutionize the adoption of federated learning in enterprises. The authors present two contributions: first, they demonstrate how smart contract-based blockchains can serve as natural communication channels for federated learning. Second, they show how an intuitive measure of each agent's contribution can be integrated into the training and reward process. The paper highlights that using smart contracts on blockchains offers several advantages for federated learning. It ensures fairness, transparency, and universality in incentivization schemes while preserving data privacy. The proposed approach has potential applications in compliance and anti-money laundering cases where multiple sensitive datasets need reconciliation and fraud/anomaly detection models can enhance audits and investigations. The authors are affiliated with Onyx Engineering at J.P.Morgan Chase & Co in India. They emphasize that their approach leverages a unique structured ledger with potential privacy layers to guarantee security, accuracy, and correctness in distributing the model's parameters. However, they acknowledge that current blockchains require privacy layers to protect communications between participants during validation. In conclusion, this paper presents a detailed exploration of how smart contract-based blockchains can facilitate federated learning while addressing challenges related to incentivization. The proposed approach has significant potential to transform the adoption of federated learning in enterprises by providing a fair and transparent reward system for contributors while preserving data privacy.