Hierarchical Learning of Conjunctive Concepts in Spiking Neural Networks

The temporal correlation hypothesis proposes that synchronous activity in different regions ofthe brain describes integral entities (von der Malsburg, 1981; Singer and Gray, 1995). Thistemporal binding approach is a possible solution to the longstanding binding problem ofrepresenting composite objects (Rosenblatt, 1961). To complement the dynamic nature oftemporal binding, a recruitment learning method has been proposed for providing long-termstorage (Feldman, 1982; Valiant, 1994). We improve the recruitment method to use a morebiologically realistic and computationally powerful spiking neuron model.However, using continuous-time spiking neurons and brain-like connectivity assumptionsposes new problems in hierarchical recruitment. First, we propose timing parameterconstraints for recruitment over asymmetrically connected delay lines. We verify theseconstraints using simulations. These constraints are useful for both building abstract networksand providing insight into bio-mechanisms that ensure signal integrity in the brain. As asecond problem, we calculate the required feedforward excitatory and lateral inhibitoryconnection densities for stable propagation of activity in hierarchical structures of thenetwork. We give analytic solutions using a stochastic population model of a simplifiedlayered network. Our approach is independent of the network size, but depends on lateralinhibition and noisy feedforward delays.