[1J3-J-2] Machine learning: bayesian models

[1J3-J-2-03] Bayesian Estimation for Spatial distribution using Low precision Sensors in Multi-Environment

〇Masato Ota¹, Ryo Hanafusa¹, Takeshi Okadome¹ (1. Kwansei Gakuin University)

Keywords:Multi-task learning, Bayesian estimation, Spatial distribution

The method proposed in this paper enables us to estimate the spatial distribution of physical quantities in multiple geographical regions, where many low-precision sensors are densely placed and a small number of (or no) high-precision sensors are positioned. For a region that has high-precision sensors, the method determines the biases of the low-precision sensors placed in the regions accurately using the values of the high-precision sensors and it corrects the values of the low-precision sensors precisely. Furthermore, the method divides the regions into the clusters using the sensor data similarity as a similarity measure. Then, for a region that has no high-precision sensor, it estimates the spatial distribution of physical quantities in the region by a novel multi-task learning that transfers the regional shared-information in the cluster which the region belongs to. Some experiments show that the method estimates the spatial distribution of physical quantities accurately.