Abstract:
The choice of transfer functions may strongly influence complexity and performance of neural networks. Although sigmoidal transfer functions are the most common, there is no /a priori/ reason why models based on such functions should always provide optimal decision borders. A large number of alternative transfer functions have been described in the literature. A taxonomy of activation and output functions is proposed, and advantages of various non-local and local neural transfer functions are discussed. Several lesser lesser known types of transfer functions, and new combinations of activation/output functions are described. Universal transfer functions, parametrized to change from localized to delocalized type, are of greatest interest. Other types of neural transfer functions discussed here include functions with activations based on non-Euclidean distance measures, bicentral functions, formed from products or linear combinations of pairs of sigmoids, and extensions of such functions making rotations of localized decision borders in highly dimensional spaces practical. Nonlinear input preprocessing techniques are briefly described, offering an alternative way to change the shapes of decision borders.

The all-digital ring-wedge detector applied to automatic object recognition
by Berfanger, David M., PhD
THE UNIVERSITY OF ROCHESTER, 2000, 162 pages

Queuing network construction using artificial neural networks
by Chambers, Mark Andrew, PhD
THE OHIO STATE UNIVERSITY, 2000, 291 pages

Nondestructive integrity analysis of multi-layer portland cement mortar slabs using spectral analysis of surface wave method
by Cho, Young Sang, PhD
POLYTECHNIC UNIVERSITY, 2000, 242 pages

Towards a genetics-based adaptive agent to support flight testing
by Cribbs, Henry Brown, III, PhD
THE UNIVERSITY OF ALABAMA, 2000, 132 pages

Embedded radial basis function networks to compensate for modeling uncertainty of nonlinear dynamic systems
by Gan, Chengyu, PhD
UNIVERSITY OF MASSACHUSETTS, 2000, 76 pages

Neural networks trained with electrophysiological data for lesion targeting in pallidotomy
by Hamilton, Jennifer Lynn, PhD
RUTGERS THE STATE UNIVERSITY OF NEW JERSEY - NEW BRUNSWICK, 2000, 182 pages

Prediction of top oil temperature for transformers
by He, Qing, PhD
ARIZONA STATE UNIVERSITY, 2000, 88 pages

The design of a simulation-based framework for the development of solution approaches in multidisciplinary design optimization
by Hulme, Kevin Francis, PhD
STATE UNIVERSITY OF NEW YORK AT BUFFALO, 2000, 152 pages

3-D defect profile reconstruction from magnetic flux leakage signatures using wavelet basis function neural networks
by Hwang, Kyungtae, PhD
IOWA STATE UNIVERSITY, 2000, 118 pages

Neural networks for function approximation and control system design
by Lavretsky, Eugene, PhD
THE CLAREMONT GRADUATE UNIVERSITY, 2000, 149 pages

A neural network-based methodology for generating spectrum-compatible earthquake accelerograms
by Lin, Chu-Chieh Jay, PhD
UNIVERSITY OF ILLINOIS AT URBANA-CHAMPAIGN, 2000, 183 pages

Automatic control of pH using an artificial intelligence approach
by Martinez, Carlos David, MS
UNIVERSITY OF LOWELL, 2000, 128 pages

Control of a nonlinear CSTR model by gain scheduling of controller tuning
by Mattingly, Richard Alan, MEng
UNIVERSITY OF LOUISVILLE, 2000, 53 pages

Sensorless position estimation for switched reluctance motors using artificial neural networks
by Mese, Erkan, PhD
RENSSELAER POLYTECHNIC INSTITUTE, 2000, 180 pages

Landscape classification and parameterization based on multiscale remote sensing and site data
by Muchoney, Douglas Michael, PhD
BOSTON UNIVERSITY, 2000, 289 pages

Estimation of mobile speed and average received power with application to corner detection and handoff
by Narasimhan, Ravi K., PhD
STANFORD UNIVERSITY, 2000, 145 pages

Forecasting macroeconomic models with artificial neural networks: An empirical investigation into the foundation for an intelligent forecasting system
by Nguyen, Dat-Dao, PhD
CONCORDIA UNIVERSITY (CANADA), 2000, 220 pages

Random iterated neural networks: Properties, evolutionary design and applications
by Nino, Luis Fernando, PhD
THE UNIVERSITY OF MEMPHIS, 2000, 91 pages