Multiparty Secure Broad Learning System for Privacy Preserving.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Multiparty learning integrates data from multiple sources to enhance performance.
• Existing privacy-preserving machine learning (PPML) methods struggle to balance security, accuracy, efficiency, and scope.
• There is a need for advanced PPML techniques that address these limitations.

Purpose of the Study:

• To propose a new PPML method, the multiparty secure broad learning system (MSBLS).
• To address the limitations of current PPML methods in simultaneously meeting security, accuracy, efficiency, and application scope requirements.
• To combine secure multiparty computing with neural networks for enhanced privacy and performance.

Main Methods:

• The proposed MSBLS method utilizes an interactive protocol and random mapping for secure feature generation.
• It employs efficient broad learning to train a neural network classifier.
• Security analysis of the MSBLS method is derived.

Main Results:

• The MSBLS method ensures that model accuracy is not reduced by encryption.
• The method demonstrates very fast calculation speeds.
• Experimental validation on three classical datasets confirms the method's effectiveness.

Conclusions:

• The MSBLS method represents a novel approach in privacy computing, uniquely combining secure multiparty computing and neural networks.
• This method offers a promising solution for privacy-preserving multiparty learning without sacrificing performance.
• The findings suggest MSBLS can effectively improve learning performance while maintaining stringent privacy requirements.