ELSC cordially invites you to the lecture given by:
Prof. Dmitri Chklovskii
Flatiron Institute, Simons Foundation.
Neuroscience Institute, NYU Langone Medical Center.
On the topic of:
“A visual motion detector: From the connectome to a theory of transformation learning”
The lecture will be held on Sunday, December 29th, at 11:00
at ELSC: Room 2004, Goodman Brain Sciences Building
Edmond J. Safra Campus
Light refreshments at 10:45
Host: Prof. Yonatan Loewenstein
Learning to detect content-independent transformations from data is one of the central problems in biological and artificial intelligence. An example of such problem is unsupervised learning of a visual motion detector from pairs of consecutive video frames. Here, by optimizing a principled objective function, we derive an unsupervised algorithm that maps onto a biological plausible neural network. When trained on video frames, the neural network recapitulates the reconstructed connectome of the fly motion detector. In particular, local motion detectors combine information from at least three adjacent pixels, something that contradicts the celebrated Hassenstein-Reichardt model.