Implementation of a large-scale optical neural network by use of a coaxial lenslet array for interconnection

Abstract

Optical implementation of a large-scale neural network with 32 x 32 neurons is reported. The experimental setup is described, error caused by limited precision of hardware is analyzed, and experimental results are presented.

General-purpose acousto-optic connectionist processor

“…This enabled us to confidently implement an HT on the system and utilize the increased vector update rate of our AO SLM. No processing capability is lost in going from a 2-D ͑LCD͒ to a 1-D ͑AO͒ input representation ͑un-less, for example, one employs a lenslet array for interconnection 10 ͒ since a 2-D representation requires the input vector to be replicated on each row.…”

“…Research has shown that imaging aberrations and light detection issues severely restrict the number of neurons available in the proposed architectures. [6][7][8][9] Recent experiments by Gao et al 10 use lenslet arrays in a coaxial arrangement to successfully implement an optical neural network with 32ϫ32 neurons. Mori et al 11 have an optical matrix-vector multiplier that uses 2-D ͑two-dimensional͒ structures ͓liquid crystal display ͑LCD͒ panels͔ to implement a two layer network.…”

Suppression of noise in optical associative memories by real-time thresholding

“…This problem becomes prominent as the data sizes grow. However, the inherent parallelism of optics has been proved to be highly promising in establishing multiple interconnections [2][3][4][5][6], making it a good candidate for associative memory implementation.…”

“…In Hopfield's original proposal, there are negative interconnection weights, producing a challenging task of performing subtractions optically, which was addressed by introducing additional interconnection channels or additional interconnection steps [2,6]. These overheads can be eliminated by adding offsets to the interconnection weights to make them all positive.…”

“…Therefore, the absence of subtraction enables a single step of optical interconnection to obtain the intermediate neurons. In previous works, photorefractive holograms [3,5] or lenslet arrays [4,6] were used to make the parallel interconnections. Here, computer-generated holograms (CGHs) are chosen for this function.…”