TALLINN JUNE 20 - 23 / 2011

Learning in and from humans: Recalibration makes (the) perfect sense

Marc O. Ernst

Bielefeld University


Cognitive Systems and Robotics have become a very active field of research in the recent years and one of the goals is to learn from the human cognitive abilities in order to transfer some of its cognitive skills into technical systems.  Humans use all their senses to construct a reliable percept representing the world with which they interact. The view I take is that in many aspects of behaviour, motor actions and multisensory processing are inseparably linked and therefore have to be studied in a closed action/perception loop. I believe that human perception and action is tailored to the statistics of the natural environment and when the environment changes our perceptions will follow these changes through the process of adaptation minimizing potential costs during interaction. In the neural processing such statistics will represent itself in probability distributions. That is, I follow Hermann von Helmholtz in the belief that human perception is a problem of inference, for which the sensory data are often not sufficient to uniquely determine the percept. Hence, prior knowledge has to be used to constrain the process of inference from ambiguous sensory signals. A principled way to describe the combination of prior knowledge with sensory data in a probabilistic way is the Bayesian Framework. Therefore, we regularly use this Bayesian Framework to construct “ideal observer” models—models that use the available information in the most optimal way, provided some task and cost function. On the one hand, these models can then be used as a benchmark against which human performance can be tested. On the other hand, these models can be formalized to scale to artificial systems. In this talk I will focus on the some of the remarkable learning capabilities of humans and I will provide examples of how our perceptions are tuned to the statistical regularities of an ever-changing environment.



Marc Ernst studied Physics in Heidelberg and Frankfurt/Main. In 2000 he received his Ph.D. degree at the Max Planck Institute for Biological Cybernetics, Tübingen, Germany, for investigations on human visuomotor behavior. For this work he was awarded the Attempto-Prize (2000) from the University of Tübingen and the Otto-Hahn-Medaille (2001) from the Max Planck Society. Starting in 2000, he spent almost 2 years as a postdoc at the University of California, Berkeley working with Prof. Martin Banks on psychophysical experiments and computational models investigating the integration of visual-haptic information. End of 2001, Marc Ernst returned to the Max Planck Institute and became principle investigator of the Sensorimotor Lab in the Department of Prof. Heinrich Bülthoff. Beginning 2007 he then became leader of the Independent Max Planck Research Group on Human Multisenory Perception and Action. Starting in 2011 he will be full professor at the University of Bielefeld, Germany leading the Department of Cognitive neuroscience. Marc Ernst  is interested in human multimodal perception, sensorimotor integration and men-machine interaction. His group participates in several international collaborative grants, including the EU Projects "ImmerSence" and "THE", investigating human-human and human-machine interaction, and a HFSP project focusing on perceptual learning. Furthermore, Marc Ernst was coordinating the 6th Framework IST European Project "CyberWalk", which developed an omnidirectional treadmill in order to enable natural free walking through Virtual Environments. Marc Ernst is President and founding member of the Eurohaptics Society and Vice Chair of the IEEE Technical Committee on Haptics. He is member of IEEE, the Robotics and Automation Society, and the Vision Science Society. Furthermore, Marc Ernst is member of the Advisory Council of the International Association for the Study of Attention and Performance and of the newly founded excellence Center for Integrative Neuroscience in Tübingen.