April 1, 2003
Tuesday - 4:00 pm in Swain West 238
Speaker: Dr. Dezhe Jin, Massachusetts Institute of Technology
Title: Fast computations with spike sequence attractors in neural networks
Abstract:
Brain consists of a large number of neurons that are intricately connected with each other. Driven by sensory inputs, the neurons behave much like pulse-coupled oscillators: The sensory inputs cause the membrane potentials of the neurons to rise and spike; the spikes are sent to other connected neurons, inducing rapid changes of their membrane potentials. How does such spiking dynamics of the neural networks underlie the computations that the brain must perform? In this talk I will discuss the use of dynamical attractors for computations in neural networks. The talk consists of three parts. First, I will show that stable, periodic, and precisely timed spike sequences are the dynamical attractors of a large class of spiking neural networks. Convergence to these attractors is fast, often within a few number of transient spikes. Second, I will demonstrate how computations such as maximum input selection (or "winner-take-all computation") can be done with these spike sequence attractors. Finally, I will discuss the possible connections between the spike sequence attractors and biological functions such as olfaction in insects.