[Paper Review] Blockchain-based Monitoring for Poison Attack Detection in Decentralized Federated Learning

Ranwa Al Mallah, David López|arXiv (Cornell University)|Sep 30, 2022

Privacy-Preserving Technologies in Data4 citations

TL;DR

This paper proposes a blockchain-based, decoupled monitoring and detection framework for poisoning attack defense in decentralized federated learning. By parallelizing monitoring (via minersMON) and training/detection (via minersFL), the approach reduces end-to-end latency, enhances scalability and robustness, and ensures reliable model aggregation even under multi-attacker scenarios with minimal performance degradation.

ABSTRACT

Federated Learning (FL) is a machine learning technique that addresses the privacy challenges in terms of access rights of local datasets by enabling the training of a model across nodes holding their data samples locally. To achieve decentralized federated learning, blockchain-based FL was proposed as a distributed FL architecture. In decentralized FL, the chief is eliminated from the learning process as workers collaborate between each other to train the global model. Decentralized FL applications need to account for the additional delay incurred by blockchain-based FL deployments. Particularly in this setting, to detect targeted/untargeted poisoning attacks, we investigate the end-to-end learning completion latency of a realistic decentralized FL process protected against poisoning attacks. We propose a technique which consists in decoupling the monitoring phase from the detection phase in defenses against poisoning attacks in a decentralized federated learning deployment that aim at monitoring the behavior of the workers. We demonstrate that our proposed blockchain-based monitoring improved network scalability, robustness and time efficiency. The parallelization of operations results in minimized latency over the end-to-end communication, computation, and consensus delays incurred during the FL and blockchain operations.

Motivation & Objective

• Address the high end-to-end latency caused by integrating poisoning attack detection in decentralized federated learning (FL) with blockchain.
• Overcome the bottleneck of centralized coordination in traditional FL by eliminating the chief node and enabling peer-to-peer collaboration.
• Improve system robustness and scalability by decoupling monitoring and detection phases in a blockchain-secured FL architecture.
• Ensure efficient, secure, and privacy-preserving model training in decentralized environments with resource-constrained devices like IoT and mobile nodes.
• Enable dynamic, real-time assessment of worker behavior to filter out malicious nodes without disrupting the training pipeline.

Proposed method

• Decouple the monitoring phase (behavior tracking of workers) from the detection and training phases using dedicated blockchain miners (minersMON and minersFL).
• Use blockchain as an immutable, trustless ledger to record and verify local model updates and worker behavior patterns.
• Assign minersMON to continuously monitor worker behavior across iterations and maintain a filtered, reliable pool of nodes.
• Allow minersFL to independently select from the pre-validated reliable set for model aggregation, avoiding delays from real-time monitoring.
• Leverage parallel execution of monitoring and training operations to minimize communication, computation, and consensus delays.
• Implement a consensus mechanism to append verified global model updates to the blockchain after aggregation, ensuring data integrity and transparency.

Experimental results

Research questions

• RQ1 How can end-to-end latency in decentralized federated learning be minimized when integrating real-time poisoning attack detection?
• RQ2 What is the impact of decoupling monitoring and detection phases on system scalability and robustness in a blockchain-secured FL environment?
• RQ3 How does the proposed framework perform under multiple poisoning attackers, and does it maintain low latency and high accuracy?
• RQ4 Can the proposed approach be efficiently deployed on resource-constrained devices such as mobile phones and IoT nodes?
• RQ5 To what extent does parallelization of monitoring and training operations reduce overall system delay compared to sequential, chief-dependent models?

Key findings

• The proposed decoupled monitoring and detection framework significantly reduces end-to-end learning completion latency by enabling parallel execution of monitoring and training tasks.
• The system maintains low convergence time even under multiple attacker conditions, demonstrating robustness and scalability across varying network sizes (3, 6, and 9 workers).
• Latency remains stable and minimal as the number of workers increases, due to the transparent separation of monitoring and training responsibilities.
• The technique is agnostic to the number of attackers, showing consistent performance with one or two attackers in the system.
• The blockchain-based design enables secure, privacy-preserving, and transparent logging of model updates and worker behavior, supporting dynamic trust assessment.
• The framework is deployable on low-resource devices such as mobile and IoT nodes, with potential for energy and bandwidth optimization through careful miner allocation.