PhD in Generating artificial EEG by combining AI and cognitive models (1.0 FTE)

We invite applications for a fully-funded four-year PhD position in the Jantina Tammes School of the University of Groningen, The Netherlands. This PhD position is a collaboration between the Cognitive Modeling group (Marieke van Vugt, https://mkvanvugt.wordpress.com ) and the department of Neurology of the University Medical Center Groningen (Natasha Maurits, https://www.clinicalneuroengineering.com & Joukje van der Naalt, https://www.rug.nl/staff/j.van.der.naalt/ ).

When clinical or cognitive questions are addressed by using deep learning or machine learning on the basis of electroencephalography (EEG) data, the availability of limited EEG data presents a significant challenge. There are at least two types of limitations on data. Firstly, it may not be possible to find enough participants due to the rarity of the disease or challenging inclusion criteria. Secondly, it may be too time-consuming, too difficult or too exhausting, to obtain enough data for within-participant classification or prediction.

A solution that has been proposed to solve the limited data problem is Generative Adversarial Networks (GANs). GANs provide an unsupervised learning method that was applied to the generation of synthetic EEG data for the first time in 2018. However, GANs have some limitations. As the GAN can only sample from a limited dataset, i.e., the real data that it is presented with, it cannot take into account the possibilities of the entire space of potential real EEG data. This limitation is especially important when the real EEG dataset is small, or obtained from a nonrepresentative sample of the population.

To improve upon current methods, in this project knowledge about underlying cognitive processes will be taken into account in the generation of artificial EEG signals to be presented to the discriminator part of the GAN network. This cognitive component will be derived from the Adaptive Control of Thought—Rational (ACT-R) cognitive architecture. ACT-R is a cognitive architecture that has been used for decades to model and predict behavior on a wide range of cognitive tasks including attentional blink tasks, complex working memory tasks, associative memory tasks and more. Moreover, it has been linked to brain activity through EEG and fMRI.

The position is one of a large number of new PhD positions embedded in the new interdisciplinary schools of the University of Groningen. Consequently, the student will have ample opportunity to interact with other PhD students and staff who are interested in pursuing interdisciplinary research on the intersection between AI, psychology, neuroscience and medicine.