John V. Arthur
Learning: Hippocampus & Olfactory System
|
 |
|||
My research goals reflect those of the lab as a whole—both to decipher the brain's computational principles and to apply such principles to engineer brainlike computers. Specifically, I am interested in the role of spike timing in encoding information and controlling synaptic plasticity. I have chosen to study these issues in archicortex (three layered), which is simpler than neocortex (six layered). I am modeling the hippocampus, which forms new memories, as well as the olfactory cortex, which recognizes familiar odors. My Silicon Hippocampus, with 1024 pyramidal neurons, each with 21 plastic synapses, learns and recalls patterns. Synapses from excitable neurons to their lethargic peers are strengthened, resulting in synchronous activation of a stored pattern despite significant variation among its constituent neurons. This synchrony is enhanced by the chip's inhibitory interneurons, which generate rhythmic activity, as shown by this movie.  Silicon Hippocampus USB2.0 Board communicate with a computer, enabling it to stimulate on-chip synapses, record silicon neuron spikes, control analog bias voltages (using analog-to-digital converters), read and write binary synaptic weights, and control a scanner to record the continuous activity of individual neurons and synapses. Further, the board includes RAM allowing arbitrary connectivity among its neurons. I embedded the Silicon Hippocampus Chip in a custom PC Board (right). I designed the board to communicate with a computer. Using USB, the computer can stimulate on-chip synapses, record silicon neuron spikes, control analog bias voltages (using analog-to-digital converters), read and write binary synaptic weights, and control a scanner to record the continuous activity of individual neurons and synapses. Currently, I am using two Hippocampus Chips to construct a novel model of episodic memory, consistent with anatomical and physiological data. This neuromorphic model will learn and recall pattern sequences, akin to how animals learn and recall sequences of locations in navigational tasks. My Silicon Olfactory System will consist of an antennal lobe (or olfactory bulb) chip and a mushroom body (or piriform cortex) chip. Work at Gilles Laurent's lab at Caltech has revealed how the antennal lobe recodes similar odor-evoked spatial patterns, reducing their overlap by augmenting the spatial code with a temporal one. And how the mushroom body recognizes these spatiotempral patterns as familiar odors. This system builds on my Silicon Hippocampus, as the underlying mechanisms are similar to those found in the hippocampus. In another project, I am designing neurons for Neurogrid. These two-compartment neurons will express tunable ionic channels that will endow them with various properties, such as spike-frequency adaptation, bursting, and post-inhibitory rebound.
Links |
