30 September 2003 On the using of CGH for artificial neurons interconnection
Author Affiliations +
One of the most promising implementation of artificial neural networks is optoelectronic implementation. Optical interconnections are useful for neural networks as far as one can take advantage of the special potential of 3D connection through free space. In hardware implementations of neural networks, the weights values will be materialized in a technological process during which various errors may occur, so that the resulting network will use more or less deviated weights. This paper studies several aspects concerning the optical interconnection of artificial neurons. The authors analyze the problems involved by using computer-generated holograms (CGH) for these interconnections and some methods of designing such diffractive elements. The authors also analyze the error sources and the consequences caused by random deviations of the neurons interconnection weights from the accurately computed values. The theoretical considerations are illustrated by designing an auto associative memory built for graphic pattern recognition. Neurons interconnections are to be implemented optically by computer generated holograms (CGH). The network functioning was simulated on computer and the paper presents also the results of simulations on a data set and a CGH layout for neuron interconnections.
© (2003) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Ioan Ileana, Ioan Ileana, Corneliu Ovidiu Iancu, Corneliu Ovidiu Iancu, Emilian Ceuca, Emilian Ceuca, Corina Rotar, Corina Rotar, "On the using of CGH for artificial neurons interconnection", Proc. SPIE 5227, Advanced Topics in Optoelectronics, Microelectronics, and Nanotechnologies, (30 September 2003); doi: 10.1117/12.519808; https://doi.org/10.1117/12.519808


Back to Top