To analyse and select potentially useful monitoring technologies to be integrated into the habitat, as well as wearable devices, in order to generate more accurate information about the individual and the actions carried out by him/her.
Develop ML models for automatic state labelling. Evaluate and select potentially useful databases for the algorithms to be developed. Develop a list of possible states and anomalies that may occur in the habitat. Carry out experiments with wearable devices, manually labelling the data generated by the simulation of listed actions. Investigate and develop ML algorithms that by means of supervised data allow an automatically labelled solution to be given.
Develop complete monitoring system. Design and develop the hardware and software that will make use of the selected sensors together with the wearable devices, enabling global data collection.
Cloud interoperability. A dynamic database architecture will be implemented to host the data generated by the different devices integrated in the home.
Develop dashboard. Develop a dashboard-type web platform that allows the visualisation of graphs on the data stored in the cloud, to facilitate its interpretation.
AI development for context interpretation. Select and evaluate the different databases with application to the new algorithm to be developed. Evaluate the application of the different types of ML algorithms capable of carrying out context interpretation, and which, together with the automatic labelling of OE2 states, is capable of improving the detection of Elioth+ML anomalies and also adding the novel capacity to interpret them. To this end, several techniques will be evaluated, such as: convolutional and variational autoencoders, deep feedback networks, approximators and uniform latent space projectors (such as UMAP).
Develop a visualisation and interpretation tool. The tools for visualisation and interpretation of latent spaces obtained through deep learning networks (Deep Learning), will eventually allow a proper (less heuristic) and simpler interpretation of situations, pattern tracking and anomaly detection by an expert in the analysis phase.
Validate algorithms. Validate the designed Machine Learning models with real data acquired with the improved and adapted ubIAsist platform.
Finding the optimal monitoring system configurations. The AI tools developed and the data generated with the complete monitoring system will allow an exploration of the quality of the results obtained if only a subset of the monitoring system’s sensors is used. In this way, it will be possible to compare each configuration from the point of view of efficiency and economic cost, and to select the most appropriate ones.
Región de Murcia
Ris 3 Mur