Boosting Intelligent Data Analysis in Smart Sensors by Integrating Knowledge and Machine Learning

Area of Science:

• Artificial Intelligence
• Machine Learning
• Sensor Technology

Background:

• Smart sensor devices are often resource-constrained and application-specific, limiting their data analysis capabilities.
• Traditional machine learning requires large datasets, which are not always available for sensor applications.

Purpose of the Study:

• To propose a hybrid neural architecture for intelligent data analysis in smart sensor devices.
• To enhance machine learning efficacy on resource-constrained hardware with limited training data.

Main Methods:

• Developed a hybrid architecture with two interacting modules: a knowledge sub-network (using L-neurons) and a conventional neural sub-network.
• Implemented knowledge in Conjunctive Normal Form within the knowledge sub-network.
• Utilized classical backpropagation for learning within the L-neurons.

Main Results:

• The hybrid structure effectively combines prior knowledge with learning from examples.
• Achieved high recognition performance even with highly limited training datasets.
• Demonstrated the architecture's ability to repair its own knowledge through learning.

Conclusions:

• The proposed hybrid neural architecture boosts intelligent data analysis in smart sensors.
• It enables successful machine learning execution with minimal data and compact hardware.
• The architecture offers a robust solution for diverse sensor applications with varying data availability.