Neural computation by time concentration.

The general problem of recognizing the pattern in a time-dependent signal is discussed and an analog model neural network which can solve this problem is presented. The networks use organized delays to collectively focus stimuli sequence information to a neural state at a future time, in a way analogous to the collective focusing of spatial information to a particular neural state in associative memories. The computational capabilities of the circuit are demonstrated on tasks similar to those necessary for the recognition of words in a continuous stream of speech. The organization of the connections and delays of the network can be understood from consideration of an energy function which is being minimized as the circuit computes. Mechanisms are known in neurobiology for the generation of delays on the requisite time scale.