Distributed semi-supervised learning algorithm based on extreme learning machine over networks using event-triggered...

Area of Science:

• Machine Learning
• Distributed Systems
• Communication Networks

Background:

• Distributed semi-supervised learning (DSSL) presents challenges for traditional algorithms due to data distribution across networks.
• Extreme Learning Machines (ELM) offer a potential framework for efficient learning but require adaptation for distributed and semi-supervised settings.
• Event-triggered (ET) communication schemes can optimize network resource usage by reducing data transmission frequency.

Purpose of the Study:

• To propose a novel distributed semi-supervised learning (DSSL) algorithm leveraging Extreme Learning Machines (ELM) and an event-triggered (ET) communication scheme.
• To address the limitations of traditional DSSL algorithms in handling distributed datasets.
• To enhance the efficiency of DSSL by minimizing communication frequency through an ET strategy.

Main Methods:

• The proposed algorithm, ET-DSS-ELM, integrates semi-supervised ELM (SS-ELM) for local initial value computation.
• A zero gradient sum (ZGS) distributed optimization strategy is employed for global optimization.
• An event-triggered (ET) communication scheme is implemented, where nodes transmit updates only upon event occurrence.

Main Results:

• The ET-DSS-ELM algorithm effectively performs DSSL by combining local and global optimization strategies.
• The ET scheme significantly reduces communication times compared to traditional methods.
• Simulations demonstrate the efficiency and effectiveness of the proposed ET-DSS-ELM algorithm.

Conclusions:

• The ET-DSS-ELM algorithm provides an efficient solution for DSSL problems over communication networks.
• The integration of SS-ELM, ZGS, and ET schemes optimizes learning performance and conserves network resources.
• The algorithm's convergence is mathematically guaranteed, ensuring reliable performance.