GENIE: Foundations of Generative Calibration for Low-cost Pervasive Light Sensing

Off-the-shelf, low-cost and miniaturized light sensors support several of our everyday applications running in smart, wearable and IoT devices. Key core and technical challenges related to their poor data quality, high variability and instability of measurements prevents their usage in critical applications that require high precision and accuracy. Existing calibration methods to overcome these issues do not perform well in practice, requiring new innovative methods to improve the performance of the sensors. GENIE explores the adoption of generative AI for light sensor calibration. The key idea is that to anticipate the potential characteristics (observed in the past or speculated via AI) of samples in specific scenarios before acquisition, allows for their enhancement and calibration, thereby increasing their effectiveness. GENIE aims to improve the performance of low-cost light sensors at comparable levels of reference devices.