Neurons, the major type of brain cells, communicate with each other by sending electrical signals called spikes. The propagation of such spikes along neuronal fibers is slow, resulting in noticeable delays. Usually, the existence of delays implies that something is not functioning well or disorganized, just think of traffic delays. However, there is a growing empirical evidence that propagation delays in the brain are maintained with a millisecond precision, that is, they are there on purpose.
The existence of delays is still ignored by the neuroscience community because their significance in neural computations is unclear. In addition, delays complicate modeling because systems with delays are infinite-dimensional from a purely mathematical point of view.
Dr. Izhikevich's experimental and computational studies of biological neurons and networks with delays show that infinite dimensionality is not a nuisance, but an immense advantage that results in an unprecedented, possibly unlimited information capacity of such systems. In particular, the systems can exhibit a novel nonlinear phenomenon -- polychronization -- generation of reproducible time-locked but not synchronous spiking patterns with millisecond precision. The number of such patterns far exceeds the number of neurons in the network, and could be even greater than the number of connections in the network.
Dr. Izhikevich proposes to explore novel computational properties resulted from the infinite dimensionality of spiking networks with delays. In particular, he intends to implement a selectional computation device, which acts according to the principles of Neural Darwinism proposed in 1987 by Nobel Laureate Gerald M. Edelman, who is the director of The Neurosciences Institute.
The major intellectual merit of the proposed research program is that it explores a radically different way of performing computations using spiking neurons, whether artificial or biological. The program advances our understanding of the role of conduction delays in information processing by the brain, and it also suggests a new computation methodology that differs fundamentally from the existing methodologies of artificial neural nets.
The proposed activity strengthens interdisciplinary postdoctoral training program at The Neurosciences Institute -- a small not-for-profit research center focused to understand how the brain works. It reinforces collaboration between theoreticians and experimentalists at the institute interested in understanding the computational power of the brain. The results, including MATLAB and C codes, will be freely available to the public.
