A Federated Learning Multi-Task Scheduling Mechanism Based on Trusted Computing Sandbox.

Area of Science:

• Computer Science
• Cybersecurity
• Distributed Systems

Background:

• Federated learning (FL) combined with blockchain enables decentralized model training.
• Existing FL-blockchain schemes lack trusted supervision and protection against malicious nodes.
• Participant data privacy and computational integrity are critical concerns.

Purpose of the Study:

• To introduce a trusted computing sandbox for secure federated learning on blockchain.
• To design a multi-task scheduling mechanism for federated learning using a trusted sandbox.
• To address resource heterogeneity in decentralized federated learning.

Main Methods:

• A decentralized trusted computing sandbox is constructed as a state channel.
• Smart contracts are utilized for supervising malicious behavior within the channel.
• Deep reinforcement learning optimizes resource scheduling for heterogeneous participant nodes.

Main Results:

• The proposed mechanism ensures data privacy and reliable computation during federated learning.
• The deep reinforcement learning-based algorithm effectively optimizes resource scheduling.
• Experimental results demonstrate superior performance compared to traditional heuristic and meta-heuristic algorithms.

Conclusions:

• The trusted computing sandbox approach enhances the security and reliability of federated learning on blockchain.
• Deep reinforcement learning is effective for optimizing resource scheduling in heterogeneous decentralized systems.
• This framework provides a robust solution for secure and efficient decentralized machine learning.