A multi temporal trainable delay neural network

Neural networks have been used for analysis of temporally related patterns. Methods used include the encoding of temporal data for input to static networks, backpropagation through time, avalanche filters, recursive networks and temporal difference learning. Each of these methods attempt to learn temporal relationships through the use of varying amplification weights coupled with a constant periodic sampling of input signals. This paper presents a method of using delays rather than amplifications to encode temporal relationships directly into the network. This method improves memory usage by the network during simulation as well as reducing the required size of the network for temporal analysis