Optical Architecture for Finite Impulse Response Neural Networks

摘要

A finite impulse response neural network (FIRNN) is equivalent to a neural network whose interconnects are composed of finite impulse response (FIR) adaptive filters.~1 Some interesting applications of FIRNNs are multi-dimensional time-series prediction, real-time speech recognition~2 and system classification, an application which consists of using adaptive FIRs to model different systems and a non-linear combination of the FIRs to classify the modeled systems.