The purpose of this work package is twofold: 1) to develop modern AI for streaming and sensor-based data analysis and 2) to develop techniques and tools for the automatic creation and management of knowledge graphs. 

The development of modern AI for straming and sensor-based data will be done by 

1. Providing anomaly detection and predictions with low quality streaming data
2. Providing uncertainty quantification and explainability with streaming data
3. Enabling combinations of streaming and static data for efficient data analysis

Streaming data can be used for automation, recommendations and decision making. Often this involves predictions and anomaly detection in multivariate time series, as well as providing explanations for the conclusions drawn. IoT sensors are increasingly instrumenting the physical world, and efforts have been made to use AI for solving these tasks also in low-quality data regimes. This research area will identify robust techniques for analysis of streaming data within several domains (including telco network, industrial IoT), with a particular focus on improving interpretability for cases with multivariate time series with low quality data. 
Solving the research problems in STREAM is crucial to successfully innovate how IoT data can be fully used in anomaly detection and contribute to breakthrough in applying AI in predictive maintenance and operational availability.

Real impact of data-driven AI depends on the availability of live data of sufficient quality and quantity in an automatically discoverable format that both humans and machines can understand. DATA will investigate how the semantics of data, through automatic creation and mapping of suitable knowledge graphs, can be leveraged to scale AI models from one situation to all similar situations and how complex graph-based structures can be efficiently stored and processed.