Page 373 - Introduction to Information Optics
P. 373
358 7. Pattern Recognition with Optics
Interconnection
Weights
Input Output
Neurons Neurons
Fig. 7.3. A single-layer fully interconnected neural network.
systems are difficult to program, in the sense of programming general-purpose
digital computers. A purely optical system can be designed to perform specific
tasks (analogous to a hard-wired electronic computer), but cannot be used
when more flexibility is required. A second problem is that a system based on
Fourier optics is naturally analog, which often makes great accuracy difficult
to achieve. A third problem is that optical systems by themselves cannot easily
Interconnection
Weights
Input
putput
Signals Signals
Neuron
Fig. 7.4. A mathematical model of a neuron.
