tinyML Talks: Datasheets for Machine Learning Sensors

Datasheets for Machine Learning Sensors
Matthew Stewart
Postdoctoral Researcher
Harvard University

Machine learning (ML) sensors have revolutionized the field of sensing, enabling intelligence at the edge and granting users greater control over their data. To support the development of intelligent devices, it is crucial to document ML sensor specifications, functionalities, and limitations comprehensively. This work introduces a standardized datasheet template for ML sensors, covering essential components such as hardware, ML model, dataset attributes, end-to-end performance metrics, and environmental impact. By presenting an exemplar datasheet for our ML sensor, we delve into each section, highlighting its significance. Our objective is to demonstrate how these datasheets enhance understanding and utilization of sensor data in ML applications, offering objective measures to evaluate and compare system performance. ML sensors, accompanied by datasheets, provide improved privacy, security, transparency, explainability, auditability, and user-friendliness for ML-enabled embedded systems. We emphasize the importance of widespread datasheet standardization across the ML community to ensure responsible and effective utilization of sensor data.


Password generation