Implementation of the neocognitron on a SIMD architecture

The Neocognitron was first proposed by Professor Kunihiko Fukushima in 1980. It is an artificial neural network based on the biological model of the human eye consisting of S (simple) and C (complex) cells. These cells work in a hirarchical fashion to selectively pick up distint patterns from an object until it is recognized. This neural network is used principally for pattern recognition.

In my thesis, in 1993, I implemented a, relatively simple, handwriting recognition application using the neocognitron. The implementation was done on the MasPar; a massively parallel computer using the SIMD (Single Instruction Multiple Data) paradigm. It worked surprisingly well and as a result I presented a paper at the Intelligent Information Systems conference in New Zealand in 1994.

Here's the abstract. (Sorry I don't have a full copy but it is available here: http://ieeexplore.ieee.org/xpl/freeabs_all.jsp?arnumber=396925)
========================
Abstract
We consider the implementation of one of the most complex neural networks, the neocognitron, on a SIMD parallel machine. The neocognitron is a network with a heavy cell load and intricate interconnections among its neurons or cells. Its structure is qualitatively similar to the mammalian visual system and its algorithm exhibits data parallelism at several levels. The proposed implementation makes use of an optimal mapping strategy exploring this data parallelism and the communication network of the MasPar Mp12000 SIMD computer
========================

I'd like to now use this same technique for something more useful. What I have in mind is sign language recognition. The idea would be to have people write a letter or manage their computers using a simple web cam. I just thought this would be a cool project.