I like to think of myself as a cognitive scientist, although my background is in physics. My research spans a number of different areas in different disciplines, including Neural Computation, Computational Neuroscience, Computer Vision, Machine Learning, Visual Psychophysics, Developmental Psychology, and Developmental robotics.

Ultimately, I am driven by the desire to understand how cognitive phenomena can arise from the collective interactions of relatively simple neural elements. In particular, I investigate how the brain's networks and subsystems can self-organize their information processing to give rise to intelligent perception and action. My research focuses on building computational models of various aspects of visual perception, action, and learning, but I also complement this research by testing specific implications of computational theories in visual psychophysics experiments and by testing the usefulness of different approaches in computer vision applications. I firmly believe that by studying the organizational principles of neural information processing through computational modeling, we can further our understanding of brain function and organization and also make progress toward building a new generation of intelligent artificial information processing systems with potentially profound social and economic implications. The long term goal I am pursuing is the development of an embodied computational account of the developing human visual system, a system that autonomously learns to perceive, understand, and interact with its environment with relatively little external supervision.

My approach to the computational modeling of cognition has three important thrusts. First, I model the brain mostly at the network and systems levels, because this is where I think our understanding of the brain still has the most glaring gaps. Our ultimate goal must be to explain observed behavior in terms of neural information processing and it becomes increasingly clear that practically all cognitive phenomena depend on widely distributed networks of neural structures, which implies that trying to understand these structures in isolation may be an ill-advised endeavor. Second, I think it is important to study the brain from a developmental perspective. The complexity of the adult brain emerges during its development and it seems likely that trying to understand the mechanisms that structure the developing brain will be easier than trying to understand the final product. Clearly, the brain cannot be a tabula rasa, but equally clearly, no detailed blue print can be specified in the genome in any explicit form. So just how do nature and nurture interact to allow the human infant to learn to perceive and understand the world around it? Third, the computational models I build are often embodied, i.e. they interact with the real world through sensors and actuators. This has two reasons: a) the developing brain is not only shaped by the self-organization of its elements but also by the bodily interactions with a highly structured environment. We simply cannot understand the brain without taking into account the world in which it learns and develops and how it interacts with this world during the process; b) computational models of brain function should ultimately be able to solve the same problems that the brain solves. Only then it is plausible that the brain may indeed work this way.