Similarity matching: a new framework for neural computation
Dmitri “Mitya” Chklovskii, Ph.D.
New York, NY
Despite our extensive knowledge of the biophysical properties of neurons, there is no commonly accepted algorithmic theory of neuronal function. Here we explore the hypothesis that a neuron performs online matrix factorization of the streamed data. By starting with a matrix factorization cost function we derive an online algorithm, which can be implemented by neurons and synapses with local learning rules. We demonstrate that such network performs feature discovery and soft clustering. The derived algorithm replicates many known aspects of sensory anatomy and biophysical properties of neurons. Thus, we make a step towards an algorithmic theory of neuronal function, which should facilitate large-scale neural circuit simulations and biologically inspired computing.