Training & innovation in tensor-based AI methods for biomedical signals

Detecting an epileptic seizure or improving our understanding how the brain works based on an ultrasound recording. These are some concrete examples of potential applications of the work of DeTAIL (Delft Tensor AI Lab).

There is more and more data to process, not only in the biomedical world. This data is also increasingly multidimensional; in other words, it comes from multiple sensors that register, for example, images, sounds and other signals.

A central problem is that the mathematical tools that are currently taught as standard are cannot sufficiently deal with these kinds of complex data streams. A key mathematical technology to improve this situation is tensor computing. This methodology builds on standard mathematics and has been developed over the last 15 years.

Tensor computing exploits the fact that there are (until now hidden) correlations in the different dimensions of the signals. Tensor computing can find these correlations, making the calculation methods much faster.

DeTAIL lab will also work on methods that calculate uncertainty margins. Current AI techniques are capable of making predictions, but no margins of uncertainty are given. This is inherent in the methods used.

Besides the further fundamental development of tensor computing, DeTAIL Lab focuses on potential applications, starting in the field of biomedical signal processing. But there are also opportunities in other fields. An example is the analysis and control of autonomous driving vehicles. After all, these are also equipped with all kinds of sensors whose data must be integrated.

At present, tensor computing is not yet a standard part of the education of engineers. In addition to its research, DeTAIL Lab will ensure that tensor computing is given a place in the Delft curriculum. TU Delft will be one of the first universities to do so.

The DeTAIL Lab is part of the TU Delft AI Labs programme.

The team


This content is being blocked for you because it contains cookies. Would you like to view this content? By clicking here, you will automatically allow the use of cookies.