Comparing Facial Line Drawings with Gray-Level Images: A Case Study on PHANTOMAS
ICANN 96 Proceedings of the 1996 International Conference on Artificial Neural Networks
Biologically inspired defenses against computer viruses
IJCAI'95 Proceedings of the 14th international joint conference on Artificial intelligence - Volume 1
Data mining techniques for optimizing inventories for electronic commerce
Proceedings of the sixth ACM SIGKDD international conference on Knowledge discovery and data mining
| Hi-index | 0.00 |
Neural networks are learning devices inspired by the workings of the brain. While the brain's precise mechanisms are far from understood, we do know that it is composed of many highly connected neurons that fire in parallel and produce various activation levels in adjacent neurons. Similarly, neural networks are composed of multiple units connected by links, where each link has an associated numeric weight. At each processing step, every unit does some local computation to determine its activation level, given the input links and weights and its previous activation level. Some units serve as input units and others as output units. A network learns from a set of training examples, which specify the values for the input and output units, by adjusting the weights on the links accordingly. Designing a neural net requires determining the number of units to use, how to connect the units, what algorithm to use for learning, how to perform the local processing, what examples to use to train the network, and how to represent the examples in terms of input and output units. (See Artificial Intelligence, by Elaine Rich and Kevin Knight [McGraw Hill, 1991] or Artificial Intelligence: A Modern Approach, by Stuart Russel and Peter Norvig [Prentice Hall, 1995], for a detailed discussion of neural-network basics.)This installment of "Trends and Controveries" looks in detail at five neural-network applications. All are in use today, either as commercial products or part of a manufacturing process. Each essay describes the application of the neural net and the challenges faced in building a practical application. In the first offering, Michael Mozer describes an inexpensive neural-network chip for speech recognition. Currently used in several children's toys, this chip provides speaker-dependent and speaker-independent, limited vocabulary recognition. In the second essay, Gerald Tesauro, Jeffery Kephart, and Gregory Sorkin present a neural-network application for recognizing computer viruses. This network can recognize previously unseen computer viruses and is currently part of the IBM AntiVirus software.Next, Wolfgang Konen describes a neural-network face-recognition application, currently used for security in a bank computer center. In the fourth essay, Martin Schlang, Thomas Poppe, and Otto Gramchow describe the use of neural networks for determining the power settings and temperatures in two steelmaking processes. These applications are providing substantial time, energy, and monetary savings for the manufacturing process. Finally, Larry Yaeger describes the use of a neural network for handwritten character recognition, which is an integral part of Apple's Newton personal digital assistant.驴Craig Knoblock, Editor