Self-balancing Incremental Broad Learning System with privacy protection.

Area of Science:

• Machine Learning
• Artificial Intelligence
• Data Privacy

Background:

• Broad Learning Systems (BLS) offer efficient remodeling without retraining.
• Existing privacy solutions like Multiparty Secure Broad Learning System (MSBLS) are limited to two clients.
• Privacy-preserving incremental learning across multiple clients remains an underexplored area.

Purpose of the Study:

• To propose a novel Self-Balancing Incremental Broad Learning System (SIBLS) for privacy-preserving multi-client incremental learning.
• To address the challenge of varying data sample sizes across clients.
• To enhance security in multi-client broad learning environments.

Main Methods:

• Developed a client selection strategy to balance data sample sizes during incremental updates.
• Introduced a mediator for secure data encryption and feature mapping.
• Implemented and validated the SIBLS using MNIST, Fashion, and NORB datasets.

Main Results:

• SIBLS demonstrates comparable performance to MSBLS.
• SIBLS outperforms federated learning in terms of accuracy and running time.
• The proposed client selection and mediation strategies effectively enhance privacy and efficiency.

Conclusions:

• SIBLS provides an effective solution for privacy-preserving incremental learning in multi-client Broad Learning Systems.
• The system balances data distribution and ensures security, outperforming traditional federated learning.
• This work advances the field of privacy-preserving machine learning for distributed systems.