SC-NBTI: A Smart Contract-Based Incentive Mechanism for Federated Knowledge Sharing.

Area of Science:

• Computer Science
• Artificial Intelligence
• Information Systems

Background:

• Digital knowledge platforms generate vast unstructured data.
• Knowledge sharing is limited by privacy, communication overhead, and data ownership.
• Federated learning enables collaborative training without raw data exchange but requires incentives for participation.

Purpose of the Study:

• To develop an incentive framework for federated learning in knowledge collaboration.
• To address limitations of existing non-cooperative game-based incentive schemes.
• To motivate data holders by managing computational and communication costs.

Main Methods:

• Introduced SC-NBTI: a smart contract and Nash bargaining-based incentive framework.
• Formulated reward allocation as a cooperative game.
• Employed a heuristic algorithm for Nash bargaining solution approximation.
• Integrated probabilistic gradient sparsification for communication efficiency and privacy.

Main Results:

• SC-NBTI achieved higher accuracy (5.89%) compared to the DRL-Incentive baseline.
• The framework required fewer training rounds for convergence.
• Demonstrated effectiveness on the FMNIST image classification task.

Conclusions:

• SC-NBTI effectively incentivizes participation in federated learning for knowledge collaboration.
• The framework balances communication costs, privacy, and reward allocation.
• Offers a promising solution for secure and efficient knowledge sharing in digital environments.