Novel Techniques for Blockchain-enabled IoT Systems Leveraging Reinforcement Learning

Abstract

In the last decade, blockchain and Smart Contracts (SCs) have attracted unprecedented attention in academia and industry due to their technical innovation of providing an immutable distributed ledger with secure cryptographic consensus rules. However, to leverage the benefits of SCs in different domains, its adaptation should take into consideration the characteristics and requirements of specific applications and systems. In this thesis, we investigate the use of SCs in the Internet of Things (IoT) applications. Specifically, we identify and propose solutions to two potential issues that might arise from such integration. First, we demonstrate that because IoT monitoring requires replicated data sources that continuously submit data as transactions to the blockchain, naive integration with SCs is prohibitively expensive. Instead, the data submission should be optimized to minimize the cost while still meeting the use-case requirement of audibility and security. We propose a Reinforcement Learning (RL)-based approach to achieve such a tradeoff and show its superior performance compared to currently followed methods. On the other hand, we also demonstrate that using SCs for task-allocation in applications like service provisioning can lead to inefficient allocation decisions due to the static nature of SCs rules that aim to manage dynamic blockchain participants. We show that leveraging the ever-expanding blockchain data for online learning by means of RL provides viable and adaptive task allocation that also outperforms currently deployed techniques in terms of cost-efficiency. Overall, the problem formulations presented here, as well as their proposed solutions, contribute to the establishment of secure and intelligent decentralized IoT applications.