A hierarchical neural network that can be trained to recognize multiple sequences of temporal events is presented. The network modules are suitable for VLSI implementation. A motion detector based on a hierarchical network was simulated and tested. The simulation results show that modules can be concatenated to recognize a long temporal sequence. Results also show that modules at higher levels in the hierarchy can be trained to recognize the temporal order of the sequences presented to modules in the lower levels. The network can be used for temporal recognition of a number of long sequences. >
