|Date and Time||March 3, 2010, 10 p.m.|
|Location||Singleton Auditorium, MIT Room 46-3002|
Speaker: Christof Koch
California Institute of Technology
Abstract: In ongoing work with the neurosurgeon Itzhak Fried at UCLA, we record chronically from multiple single neurons in the medial temporal lobe (MTL) of patients with pharmacologically intractable epilepsy implanted with depth electrodes in order to localize the focus of seizure onsets. These neurons fire in a remarkably selective manner to different images of famous or familiar individuals and objects. These data supports a sparse, abstract, invariant and modality-independent representation in MTL, suggesting that the identity of individuals is encoded by a small number of neurons. I will describe these findings and estimate their sparseness using Bayes' rule, concluding that these cells bear some resemblance to a Grandmother Cell representation. Will discuss an unsupervised learning scheme based on sparse coding that gives rise to such cells. By feeding the firing activity of these cells back to an image display in < 100ms, we show that subjects can voluntarily, rapidly and differentially control the content of this image by focusing their thoughts specifically onto one out of four competing concepts associated with each of four simultaneously recorded MTL neurons. We show that subjects can rapidly, sometimes on the first trial, learn to regulate the firing rate of group of neurons deep inside their own brain, increasing the rate of some while simultaneously decreasing the spiking rate of others.
There are no comments yet