Figure 3
From: A digital nervous system aiming toward personalized IoT healthcare
Digital nervous system integration and implementation of machine learning. (a) The DNS proof-of-concept, where the sensing node (glove) state was received by the base unit/tag (in contact with person, but out-of-view during filming) and forwarded to the actuator node (ion pump) through the capacitive field of the body. The received state was further sent from the base unit to the mobile device via Bluetooth for interaction with the Hopsworks cloud-based machine learning system (steps 1–4 and a–e). (b) Accuracy of a neural network classification of the hand position as a function of the quantity of data used to train it, compared to the accuracy of an initial estimated threshold and an optimized threshold corresponding to bent fingers. (c) Accuracy of a neural network classification of the hand position as a function of the simulated drift in sensor measurements compared to the training data. The drift was applied on one finger (grey) or all fingers (orange) at the same magnitude. Photographs in part a taken by Thor Balkhed at Linköping University, and used with permission.
