Toward Accelerated Training of Parallel Support Vector Machines Based on Voronoi Diagrams.

Area of Science:

• Machine Learning
• Wireless Sensor Networks
• Data Science

Background:

• Wireless sensor networks (WSN) are crucial for Industry 4.0 and smart cities, facing challenges in processing large federated datasets.
• Machine learning in WSN requires reduced energy consumption and minimized inter-device data exchange.

Purpose of the Study:

• To introduce a novel method for accelerated training of parallel Support Vector Machines (pSVMs) tailored for WSN.
• To address energy efficiency and data exchange challenges in federated learning for WSN.

Main Methods:

• Developed a novel ensemble-based pSVM training method.
• Split training data into Voronoi regions for faster parallel SVM training.
• Compared the proposed method against single SVM and standard ensemble SVMs.

Main Results:

• The proposed pSVM method achieves comparable performance to standard approaches.
• The method effectively reduces the number of regions required for classification tasks.
• Demonstrated potential for developing energy-efficient policies in WSN.

Conclusions:

• The novel ensemble pSVM training method offers an efficient solution for WSN applications.
• This approach facilitates energy savings and reduced data communication in federated learning scenarios.
• The method is suitable for Industry 4.0 and smart city environments.