Abstract:Federated learning (FL) is a distributed machine learning paradigm in the Internet of Things (IoT), which allows multiple devices to collaboratively train models without leaking local data. In the open scenario of IoT, malicious devices can launch poisoning attacks to compromise the final model by submitting crafted gradients. Some previous studies defend against poisoning attacks by analyzing the statistical characteristics of plaintext gradients. However, plaintext gradients would expose private information to malicious FL devices or servers. To simultaneously resist poisoning attacks and preserve privacy, cryptography technology can be utilized to obfuscate the gradients in defense methods, but the private calculation of resisting poisoning attack methods will cause efficiency problems, especially imposing unaffordable overhead on resource-limited IoT devices. Therefore, resisting poisoning attacks efficiently while protecting privacy remains a challenge. This paper proposes a secure and privacy-enhanced FL (SPEFL) framework for efficient privacy-preserving and poisoning-resistant federated learning in IoT. We design an efficient secure computation protocol based on a three-server architecture to facilitate the cryptographic computation of large linear and complex nonlinear operators in the method against poisoning attacks. In SPEFL, most of the calculations are efficiently performed on the servers, which will not impose too much burden on resource-limited IoT devices. In addition, we design a security-enhanced verifiable protocol to detect the malicious behavior of the server and guarantee the correctness of FL aggregation results. Experimental and theoretical results demonstrate that SPEFL can efficiently complete FL training meanwhile guaranteeing the accuracy of the model.

A Blockchain-based Federated Learning Framework for Defending Against Poisoning Attacks in IIOT

Defending Against Poisoning Attacks in Federated Learning with Blockchain

FDSFL: Filtering Defense Strategies Toward Targeted Poisoning Attacks in IIoT-Based Federated Learning Networking System.

Secure and Efficient Decentralized Federated Learning with Data Representation Protection

BPFL: Blockchain-Based Privacy-Preserving Federated Learning against Poisoning Attack

A Blockchain-Reinforced Federated Intrusion Detection Architecture for IIoT

BlockDFL: A Blockchain-based Fully Decentralized Peer-to-Peer Federated Learning Framework

Federated Learning based on Defending Against Data Poisoning Attacks in IoT

Dependable Federated Learning for IoT Intrusion Detection Against Poisoning Attacks

DefendFL: A Privacy-Preserving Federated Learning Scheme Against Poisoning Attacks

Blockchain-Enabled Federated Learning Data Protection Aggregation Scheme With Differential Privacy and Homomorphic Encryption in IIoT

SPEFL: Efficient Security and Privacy Enhanced Federated Learning Against Poisoning Attacks

A Blockchain-based federated learning framework for secure aggregation and fair incentives

FL-PTD: A Privacy Preserving Defense Strategy Against Poisoning Attacks in Federated Learning.

Fair and Robust Federated Learning via Decentralized and Adaptive Aggregation based on Blockchain

Clean-label Poisoning Attacks on Federated Learning for IoT

Dynamic Asynchronous Anti Poisoning Federated Deep Learning with Blockchain-Based Reputation-Aware Solutions

Privacy-Preserving Asynchronous Federated Learning Framework in Distributed IoT.

Blockchain-based Monitoring for Poison Attack Detection in Decentralized Federated Learning

Blockchain-based Asynchronous Federated Learning for Internet of Things