Artificial and natural computations for sensory perception: what is the link?
Attendance type(s): In Person
Event Dates: 06—13 Jun 2020
Recent advances in machine learning have yielded deep neural network models with impressive performance in artificial perception, based in part on principles established by experimental studies of sensory neuroscience. This is an exciting time for research in this field, as such models provide quantitative theories for the key computations that give rise to perception in biological sensory systems. However, unlike brains, current deep neural network models are often brittle and do not generalize well to new situations. At the same time, many structural and functional aspects of real sensory systems appear to be incompatible with the very simple design of neural networks used in artificial intelligence (AI).
In this school, we will discuss analogies and discrepancies between sensory system computations and artificial intelligence models and how the two fields can interact more to advance each other. A set of international lectures will provide an overview of modern algorithms in artificial intelligence and recent breakthroughs in the neural mechanisms of sensory perception. The students will also learn how to practically implement AI tools, such as deep neural networks, and we will also discuss the ethical context of neuro-inspired computational technologies.
Matthias Bethge, University of Tübingen, Germany
Michael Brecht, Bernstein Center Berlin, Humboldt-Universität Berlin, Germany
Na Ji, University of California Berkeley, USA
Jennifer Linden, The Ear Institute, Faculty of Brain Sciences, UCL, UK
McKenzie Mathis, EPFL, Switzerland (2020) & Rowland Institute at Harvard University, USA
Alexander Mathis, Harvard University, USA & EPFL (May 2020), Switzerland
Josh McDermott, Department of Brain and Cognitive Sciences, MIT MA, USA
Maneesh Sahani, Gatsby Computational Neuroscience Unit, UCL, UK
Edgar Y. Walker, Department of Neuroscience, Baylor College of Medicine, Houston, USA
Dan Yamins, Departments of Psychology and Computer Science, Stanford University, USA