Queuing theory for spike driven synaptic dynamics
We present a model for spike driven, stochastic dynamics of a synapse with two stable states, suited for aVLSI implementation. The stochastic nature of learning, which allows for optimal storage capacity, is due to the variability in spike emission times of pre- and post-synaptic neurons, and emerges as a result of the collective properties of the whole network. The dynamics of the single synapse is studied with the methods of queuing theory. Numerical results show that LTP and LTD are stochastically induced by the two neurons’ activity states and transition probabilities are in the range required by the theory of stochastic learning.
Unable to display preview. Download preview PDF.
- Fusi S. and Mattia M. Collective behavior of networks with linear (VLSI) integrate-and-fire neurons, submitted to Neural Computation, (1998); Mattia M., Del Giudice P., Amit D. J., Asynchronous simulation of large networks of spiking neurons and dynamical synapses, these proceedings Google Scholar
- Annunziato M., Badoni D., Fusi S., Salamon A., Analog VLSI implementation of a spike-driven stochastic dynamical synapse, these proceedings Google Scholar