Transfer Learning for Neuronal Networks Deployed on the Sensors Edge

The usage of neuronal networks to directly process measurement data recorded by a sensor setup possesses many benefits. However, outside a laboratory application the environment of a sensor must be taken into account for the training phase of the network. This is problematic since many aspects are only known after the installation and therefore are not reflected by the training data. On this account the application of transfer learning is investigated in this work in order to produce a relief for individual installations of mass-produced sensors.