Research Interest

My interest for responsible applications of artificial intelligence led me to the field of cognitive science. My research projects range from applications like an autonomous coach for training ones cognitive control in everyday's working scenarios to making neuroscience more efficient by augmenting EEG data saving resources like time and money and reserving scientists more capacity for the creative part. My current PhD project contributes directly to the mission of the Autonomous Empirical Research Group by implementing a framework for the automated research of human reinforcement learning in a closed loop without the need for human intervention.

Automated scientific discovery of human cognition

The potential of automating scientific practice is becoming increasingly recognized as a novel paradigm for scientific discovery within the natural sciences, ranging from the prediction of protein folding to the automated synthesis of novel materials. Yet, approaches to automating scientific discovery have, so far, evaded cognitive science—in part, because human behavior is noisy and results from complex cognitive dynamics that are unobservable. In my PhD, I seek to develop a closed-loop machine learning framework for automating the discovery of latent dynamics underlying human reinforcement learning. The framework integrates computational model discovery, optimal experimental design, and web-based data collection into a closed-loop system for empirical research, and provides the basis for automated behavioral laboratories. Preliminary results indicate that this framework is capable of recovering a broad class of human reinforcement learning models from noisy behavioral data. During my PhD, I will continue to pilot this system for the data-driven discovery of novel scientific models of human reinforcement learning, and explore its generalizability to explain models of human decision making. The general methodology is inspired by successful applications of automated scientific discovery in the natural sciences and has the potential to significantly accelerate the study of human cognition and behavior, while enhancing transparency and replicability of behavioral research.

EEG data augmentation

There is major potential for using electroencephalography (EEG) in brain decoding that has been untapped due to the need for large amounts of data. Advances in machine learning have mitigated this need through data augmentation techniques, such as Generative Adversarial Networks (GANs). Here, we gauged the extent to which GANs can augment EEG data to enhance classification performance. Our objectives were to determine which classifiers benefit from GAN-augmented EEG and to estimate the impact of sample sizes on GAN-enhancements. We investigated three classifiers—neural networks, support vector machines, and logistic regressions— across seven sample sizes ranging from 5 to 100 participants. GAN-augmented EEG enhanced classification for neural networks and support vector machines, but not logistic regressions. Further, GAN-enhancements diminished as sample sizes increased—suggesting it is most effective with small samples, which may facilitate research that is unable to collect large amounts of data.

Creating an adaptive and autonomous cognitive control coach

In the pursuit of long-term goals like learning new skills or advancing in career paths, individuals often face the challenge of navigating through distractions prevalent in the digital world. These distractions, promising quick rewards, can derail efforts towards achieving these aspirations, demanding significant cognitive control to resist. However, the development of this cognitive control skill is feasible through training and by integrating an artificial cognitive control coach into our daily life we can effectively foster this skill. This coach provides feedback to a person based on their training history and current cognitive state, utilizing an Actor-Critic reinforcement learning architecture. This setup proves valuable in enhancing cognitive control, even though the exact amount exerted cannot be directly measured. The coach, in this case the critic, assesses the value of the current cognitive state without knowing the exact amount of cognitive control and returns its feedback to the individual, in this case the actor. That way, a multitude of cognitive mechanisms like mental fatigue over time, the value of learning or reward-driven distractibility can be considered in that evaluation leading to an adaptive and autonomous cognitive control coach.