Design, Analysis and Validation of Biologically PlausibleComputational Models (Funded by the NSF)
One of the fundamental difficulties in studying brain function is how to abstract the computation and model the dynamical behavior of trillions of neurons. This proposal puts together a synergistic effort to advance the theory of biological plausible computation away from the local minimum created by modeling both the neuron as a static processing element (the McCulloch and Pitts model) and neural assemblies as dynamical systems with very restrictive dynamics. We seek more powerful computational models, which will exploit the time dimension in a more principled and productive manner. We center our studies in the recently proposed liquid state machine (LSM), which embodies computation principles very different from Von Neumann machines. LSMs are built from a recurrent neural network core (the liquid state (LS)) with nonconvergent dynamics combined with a static mapper readout, and have been shown to be universal computing machines in myopic functional spaces.
In this proposal we establish a plan to further test the liquid state machine principles by replacing the artificial recurrent neural network part of the LSM with a living neural network in the form of a dissociated neuron culture (DCT), effectively creating a real time hybrid (artificial-biological) computer. Presently, DCTs appear as the most realistic and efficient test bed to the theoretical predictions obtained with the LSM because of the full control of inputs and extensive probing of neurons. A set of well-defined questions was established to compare the properties of the LSM with those of a population of neurons grown in culture on a multi-electrode array. At the same time we plan to enhance our understanding of the liquid state properties advancing a more tight connection with the theory of dynamical systems. Specifically, we propose to integrate the LSM model with our studies of computation in living neural networks into a system where both model and empirical data from cultured neural networks can be systematically investigated to improve our understanding of neural computation, and determine the computational capabilities of DCTs.
20 Node Xserve Cluster
References for other models of these or similar networks:
Chao, Z. C., Bakkum, D. J., Wagenaar, D. A., & Potter, S. M. (2005). Effects of random external background stimulation on network synaptic stability after tetanization: a modeling study. Neuroinformatics, 3(3), 263-280.
Netoff, T. I., Clewley, R., Arno, S., Keck, T., & White, J. A. (2004). Epilepsy in small-world networks. J Neurosci, 24(37), 8075-8083.
Persi, E., Horn, D., Volman, V., Segev, R., & Ben-Jacob, E. (2004). Modeling of synchronized bursting events: the importance on inhomogeneity. Neural Comput, 16(12), 2577-2595.
Segev, R., Shapira, Y., Benveniste, M., & Ben-Jacob, E. (2001). Observations and modeling of synchronized bursting in two-dimensional neural networks. Physical Review E, 6401(1), -.
Volman, V., Baruchi, I., Persi, E., & Ben-Jacob, E. (2004). Generative modelling of regulated dynamical behavior in cultured neuronal networks. In Physica a-Statistical Mechanics and Its Applications (Vol. 335, pp. 249-278).