High-capacity photorefractive neural network implementing a Kohonen topological map

Abstract

We designed and built a high-capacity neural network based on volume holographic interconnections in a photorefractive crystal. We used this system to implement a Kohonen topological map. We describe and justify our optical setup and present some experimental results of self-organization in the learning database.

Photorefractive Polymers

“…Illumination with uniform intensity distribution of appropriate wavelength will erase any grating that has already been written inside the material, making them suitable for reversible and real-time holography. Many applications such as multiplexed data storage [2], holographic filters [3], neural networks [4], and updatable 3D display [5] have been demonstrated on a laboratory scale, but commercial products are not yet available mainly because of the strict requirements on material quality and the complexity of the optics required to implement these systems. Early searches for better materials had been concentrated on inorganic crystals, but after the invention of polymer PR materials, significant research efforts have directed to this class of materials as well due to the ease of processability, which is an advantage over inorganic crystals.…”