Temporal Coding of Neural Stimuli

Abstract

Contemporary artificial neural networks use various metrics to code input data and usually do not use temporal coding, unlike biological neural systems. Real neural systems operate in time and use the time to code external stimuli of various kinds to produce a uniform internal data representation that can be used for further neural computations. This paper shows how it can be done using special receptors and neurons which use the time to code external data as well as internal results of computations. If neural processes take different time, the activation time of neurons can be used to code the results of computations. Such neurons can automatically find data associated with the given inputs. In this way, we can find the most similar objects represented by the network and use them for recognition or classification tasks. Conducted research and results prove that time space, temporal coding, and temporal neurons can be used instead of data feature space, direct use of input data, and classic artificial neurons. Time and temporal coding might be an important branch for the development of future artificial neural networks inspired by biological neurons.