Optical neural networks using a new radial nonlinear neural layer

Miyata, Yoshiro; Fellows, Mike; Wagner, Kelvin; Mozer, Michael C.; Smolensky, Paul

doi:10.1117/12.983192

Cited by 1 publication

(1 citation statement)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…One example of such a radial nonlinearity is (3) 1+ Iii where the k are the net inputs to the radial slab, and that some applications, such as winner-take-all networks, and some kind of pattern classification, intrinsically require a nonlinear rotation of the hidden layer neuron inputs, (sometimes governed by attractor dynamics along the coordinate axis), yielding a hard decision-making capability. However, it appears that many other nonlinearly separable problems in logic, pattern recognition, and neural control, can be adequately and sometimes even advantageously handled with purely radial nonlinearities.31 The idea of radial nonlinear neural slabs is compared with conventional layers of point nonlinear neurons in Figure 1.…”

Section: Radial Neural Nonliiiear Slabsmentioning

confidence: 99%

<title>Optical neural networks using a new radial nonlinear neural layer</title>

et al. 1993

Self Cite

View full text Add to dashboard Cite

Radially nonlinear neurons are introduced, and back propagation learning for multilayer networks of these simple hidden units is derived and simulated. The nonlinear transformation performed by a hidden layer of radial units can be represented as a simple multiplication of the summed net input to each neuron by a single value which is only dependent on the total input to the hidden layer. This allows a simple optical implementation, in which a single modulator/detector is able to act as an entire hidden layer by multiplexing the neuron net inputs and processed outputs.

show abstract

Section: Radial Neural Nonliiiear Slabsmentioning

confidence: 99%