Real systems, such as the human body or the climate, are constantly changing due to complex cause-and-effect relationships. Similar to a film in which a plot develops. The measurement and observation data that researchers use to identify these relationships are like snapshots, describing the system’s state at a single point in time. For example, a study participant may note the irregular occurrence of pain. Similarly, a climate researcher collects images and measurement data whenever an expedition is possible.

Until now, researchers have lacked adequate methods for analysing this data. One established approach identifies causal relationships, but it only works reliably with regularly collected data. Another method maps continuous changes in the system and can classify irregular data, but cannot identify causal relationships, let alone distinguish between true and false ones.

Powerful tool for a wide range of disciplines

TU computer scientist Nicholas Tagliapietra and his team have developed a new method called 'Causal Discovery for Dynamic Timeseries' (CaDyT) that can do both. CADYT makes it possible to identify causal relationships in irregularly collected data for the first time, even for systems that change over time. To achieve this, the researchers combine approaches from causality, information theory, and dynamic systems.

CaDyT provides researchers from a wide range of disciplines with a powerful tool for better understanding complex, dynamic systems. Tagliapietra is conducting his doctoral research at the Bosch Centre for Artificial Intelligence as an external TU student, under the supervision of TU professor Kristian Kersting.