Talk: Studying and modifying the dynamics of neural networks by Larry Abbott
About the speaker: http://www.neurotheory.columbia.edu/~larry/
Comments: It’s the first time I listened to his talk, and he is really an impressive person, very smart and scientific. I haven’t started reading his book “Theoretical Neuroscience” yet, which is the text book of computational neuroscience in MIT. His experiment on neural network is quite enlightening, which hints that the chaotic world could be synchronized into an orderly output with a stimulus, no matter what inside is. The model is quite interesting and may help to explain many mechanism of neural development or diseases. I will read more before writing in details here.
Summary from “Generating Coherent Patterns of Activity from Chaotic Neural Networks” Neuron, 2009, 63, 544:
Neural circuits display complex activity patterns bothspontaneously and when responding to a stimulus orgenerating a motor output. How are these two formsof activity related? We develop a procedure calledFORCE learning for modifying synaptic strengthseither external to or within a model neural networkto change chaotic spontaneous activity into a widevariety of desired activity patterns. FORCE learningworks even though the networks we train are spontaneouslychaotic and we leave feedback loops intactand unclamped during learning. Using this approach,we construct networks that produce a wide variety ofcomplex output patterns, input-output transformationsthat require memory, multiple outputs that canbe switched by control inputs, and motor patternsmatching human motion capture data. Our resultsreproduce data on premovement activity in motorand premotor cortex, and suggest that synaptic plasticitymay be a more rapid and powerful modulator ofnetwork activity than generally appreciated.
