Robust and Data Efficient Classification for IoT based sensing

Robust and Data Efficient Classification for IoT based sensing

연구 분야: Infrastructure

학회: BuildSys '24: Proceedings of the 11th ACM International Conference on Systems for Energy-Efficient Buildings, Cities, and Transportation

For daylong indoor events classification, energy-efficient design of indoor sensors are vital. Indoor sensors are often designed to transmit collected data to remote platforms, such as cloud or edge devices, to handle resource-intensive post-processing tasks. To extend operational lifespan under power restrictions, these sensors should operate smartly. Since fluctuations in sensed values related to indoor events are typically brief, indoor sensors should suppress insignificant part and offload only critical information for remote classification. This approach can enormously conserve energy on the sensor node by minimizing number of required wireless transmissions for offloading, as radio transmissions are far more power-intensive than local processing. However, challenges like identification of key segments, balancing energy efficiency with classification accuracy, determining the appropriate data compression technique, and finally, ensuring robustness in real-world conditions must be addressed. This study will explore solutions to these challenges.