The CCN Group combines computational neuroscience, deep learning, neuroimaging and robotics to study sensorimotor systems with a focus on perceptual & motor learning, visual processing, bottom-up & top-down attention and decision making.
Perceptual & Motor learning
The human brain has the fascinating ability to learn new skills throughout life. Perceptual learning refers to lasting, practice-induced, improvements of perception skills such as differentiating similar visual patterns. Similarly, motor learning refers to lasting, practice-induced, improvements in fine motor skills. Both types of learning are highly relevant to develop real-world expertise such as reading and writing.The CCN group studies perceptual learning by developing biologically plausible neural mass models of the visual system and exposing them to learning paradigms typically employed in psychophysics experiments. In collaboration with Prof. De Weerd and his team at Maastricht University, the behavior of these models is validated against human behavior in identical paradigms. The CCN group, further, employs deep reinforcement learning to teach robotic agents fine motor control such as reaching, grasping and in-hand object manipulation. This work is carried out in collaboration with the Robotics, Artificial Intelligence and Real-Time Systems group at Technische Universität München (TUM) and the Research Center for Information Technology at the Forschungszentrum für Informatik (FZI) Karlsruhe.
Humans are incredibly visual beings. Not only is vision pivotal for both recognizing objects and localizing them in space, it also serves an important role in cognition. For instance, visual mental imagery, the fascinating phenomenon of quasi-perceptual experiences in the absence of external stimulation, has been linked to working memory, problem solving, and creativity. The CCN group both utilizes existing tools offered by computational neuroimaging as well as develops new tools to study the processes underlying visual perception and visual mental imagery. Insights regarding visual perception are subsequently used to improve deep neural networks while insights regarding mental imagery are used to advance vision-based brain-computer interfaces.
Bottom-up & Top-down attention
Only a small central region of the eye, known as the fovea, perceives incoming light with high resolution, whereas resolution decreases rapidly towards the periphery. Given the limited number of photoreceptors in the eye, this arrangement provides an optimal trade-off between resolution and coverage. High resolution information of any location in visual space can be obtained by fixating at these locations. However, not all regions are equally important. To guide eye movements preferably to important regions of a scene, the brain utilizes bottom-up and top-down attention mechanisms. The CCN group studies the topographic organization of bottom-up (saliency map) and top-down (relevance map) attention as well as their integration (priority map) by developing deep convolutional neural networks able to predict human fixation patterns as well as by studying the cortical regions involved in attention using ultra-high field functional magnetic resonance imaging.
Decision making is an integral part of many of the processes and capabilities studied by the CCN group. Perceptual learning requires decisions about whether a stimulus is present or whether two stimuli are distinct, in-hand object manipulation requires decisions regarding how to move each finger given the current state of the manipulated object and the hand and object recognition requires decisions where to look next in a visual scene based on saliency information. Given the importance of decision making in all of these contexts, the CCN group investigates neural processes underlying decision making using mean field and neural mass models.