An Evolving TinyML Compression Algorithm for IoT Environments Based on Data Eccentricity

Area of Science:

• Computer Science
• Data Science
• Machine Learning

Background:

• Internet of Things (IoT) applications generate vast amounts of sensor data, posing challenges for storage and management.
• TinyML enables machine learning on resource-constrained devices, facilitating real-time local data analysis.
• Efficient data compression is crucial for managing IoT data on small devices and optimizing wireless communication.

Purpose of the Study:

• To introduce a novel data compression algorithm for IoT environments leveraging the TinyML paradigm.
• To present the Tiny Anomaly Compressor (TAC) algorithm, based on data eccentricity, requiring no prior data distribution assumptions.

Main Methods:

• Developed the Tiny Anomaly Compressor (TAC) algorithm utilizing data eccentricity for IoT data compression.
• Conducted a comparative analysis of TAC against Swing Door Trending (SDT) and Discrete Cosine Transform (DCT) using two real-world datasets.

Main Results:

• The TAC algorithm achieved a maximum compression rate of 98.33%.
• TAC demonstrated superior performance over SDT and DCT in terms of compression error and peak signal-to-noise ratio across all tested scenarios.

Conclusions:

• The proposed TAC algorithm offers a highly effective solution for IoT data compression, particularly within the TinyML framework.
• TAC's ability to achieve high compression rates and low error makes it suitable for resource-limited IoT devices and efficient network communication.