Self-architecture-theory and experiment of biological neural network

Live neuron behavior on an electronic chip is recorded with a time-lapsed video under the microscope. By means of an image processing technique, the smallest size of intelligent biological neural networks is discovered. The dissociated chick embryonic brain neurons are experimentally placed on silicon glass plates deposited with metal oxide strips of about 10 μm width for possible electronegativity neurite guidance. The neurite growth connecting other neurons is accomplished intentionally through selective paring in time rather than mechanically following the external electronegativity guidance. The authors theoretically map to artificial neural networks (ANNs) to determine whether the connectivity patterns dictate the information processing efficiency, or vice versa. A dynamic interconnection model is based on the energy landscape, E(v_i, W_ij), which is a function of both the individual neuron output firing rates, v_i, and the pair synaptic weights, W_ij, of which a linear Hebbian rule becomes a special case, and a general convergence theorem is proven. A nonlinear backprop-like learning rule associated with slopes of the singlet sigmoidal function and the pair-correlation function is derived