Event-Driven Deep Learning for Edge Intelligence (EDL-EI).

Area of Science:

• Computer Science
• Artificial Intelligence
• Internet of Things

Background:

• Edge intelligence (EI) is crucial for low-latency, efficient, and private data processing.
• The proliferation of Internet of Things (IoT) devices generates vast amounts of data at the edge.
• Challenges exist in EI due to heterogeneous computational capabilities and distributed data sources.

Purpose of the Study:

• To propose a novel event-driven deep learning framework (EDL-EI) for edge intelligence.
• To address challenges in EI by integrating multi-sensor fusion and lightweight deep learning models.
• To demonstrate the feasibility and effectiveness of the EDL-EI framework in real-world scenarios.

Main Methods:

• Developed a novel event model using correlation analysis and multi-sensor fusion.
• Implemented a transformation method to convert sensor streams into images.
• Utilized lightweight 2-dimensional convolutional neural network (CNN) models for deep learning.
• Built an IoT-based prototype system with multiple sensors and edge devices.

Main Results:

• Achieved outstanding predictive accuracy (97.65% and 97.19%) for air quality patterns in South Korea and China.
• Demonstrated the feasibility of the EDL-EI framework through an IoT prototype system.
• Analyzed air quality changes from 2019-2020 to assess the impact of COVID-19 lockdowns.

Conclusions:

• The EDL-EI framework effectively leverages multi-sensor fusion and lightweight CNNs for accurate edge intelligence.
• The proposed framework shows significant potential for real-time edge data analytics in IoT environments.
• The study provides insights into air quality dynamics and the influence of external factors like the COVID-19 pandemic.