Artificial Neural Oscillators for Inferencing

Trent E. Lange, Jacques J. Vidal, Michael G. Dyer


Connectionist networks have been unable to perform high-level conceptual tasks because of their inability to handle the variable binding and inferencing problems. This paper proposes the use of artificial neural oscillators in a localist network to approach these problems. In this model, groups of relaxation oscillators with unique patterns of natural oscillation frequencies serve as signatures to identify the concepts bound to an oscillator "variable" Inferences are made as the frequency signatures representing variable bindings propagate across chains of phase-locking oscillators.