Optical neural networks

Article Abstract:

Optical neural networks combine optical information processing concepts and neural network theory. Digital computers are good at solving algorithmically derived problems, which the human brain solves slowly and poorly. The human brain is adept at recognizing complex patterns, generalizing, abduction, intuition, problem finding and language. Neural networks are brain-inspired processors that use a small number and a few types of structural components in the brain; the components are adapted for electronic or optical implementation. Digital computers and neural networks can be combined: neurons can perform Boolean operations. Optical neural networks can be created by adapting 'classical' neural networks, optical computing and optical components. A systematic morphology of optical neural networks is provided, and the special problems they create are discussed.

Technology, Product introduction, Implementation, Computer Design, Scientific Research, Hybrid Computers, New Technique, Models of Computation, Optical Computers

---

A statistical approach to learning and generalization in layered neural networks

Article Abstract:

The Gibbs formulation of statistical mechanics can be used to determine the probability that a layered network will correctly predict the solutions for statistically independent examples. The free energy of an ensemble of networks is equivalent to the training data's stochastic complexity. The prediction distribution's entropy measures network performance consistently. The statistical mechanics of neural networks are linked to statistical estimation methods and can be used to optimize architecture and predict learning curves.

Algorithms, Algorithm, Models, Statistical Analysis, Computer Learning, Parametric Tests

---

Similar abstracts:

• Abstracts: Neural network models of sensory integration for improved vowel recognition. An Improved Conference Circuit
• Abstracts: The current status of laser angioplasty: coronary and peripheral results
• Abstracts: Synthetic neural modeling: the "Darwin" series of recognition automata
• Abstracts: Image-enhanced estimation methods. Order statistics in digital image processing. Image understanding environments

This website is not affiliated with document authors or copyright owners. This page is provided for informational purposes only. Unintentional errors are possible.