LearnChain: Transparent and cooperative reinforcement learning on Blockchain

Hani Sami, Rabeb Mizouni, Hadi Otrok, Shakti Singh, Jamal Bentahar, Azzam Mourad

Abstract: Highlights•Perform on-chain MARL model update in the context of CTDE to empower transparency and trust.•Propose LearnChain, a novel framework for integrating DRL for multi-agents and Quorum Blockchain.•Design and divide a set of smart contracts to handle the model lifecycle starting with data handling, deployment, learning, target update, incentive mechanism, and inference.•Present a framework implemented using Solidity and Web3js and evaluated on use cases extracted from a real-world dataset.•Perform a list of analyses on the performance, quality of decisions following environment changes, and optimization required for efficiently deploying the MARL model.