View Table of ContentsCreate a BookmarkAdd Book to My BookshelfPurchase This Book Online
Skip to Book Content
Book cover image

Index

Neural Smithing: Supervised Learning in Feedforward Artificial Neural Networks
Russell D. Reed and Robert J. Marks II
Copyright © 1999 Massachusetts Institute of Technology

Previous Section Next Section

Index

G

Gain competition, 232-233
Gain scaling, 105, 114
generalization and, 133-134, 274, 283-287
hard-limiters and, 134
learning rate and, 132-133
sigmoid saturation and, 133
Game playing, 12
Gaussian cumulative distribution function, 284-285, 288, 311-314
Gaussian probability distribution function, 288-289, 311-314
Gaussian weight distribution, 100-102
Gauss-Newton method, 130, 172
Generalization
Akaike's methods and, 260-261
algorithm factors and, 157, 249-253
Bayes' rule and, 258-260
bias and, 240, 249, 256, 258, 267
complexity and, 241-242, 244-247
constructive methods and, 198-199, 268
cross-validation of, 257-258, 273, 291
data factors and, 242-248
definitions for, 239-240
early stopping for, 265-266
factors affecting, 240-256
gain scaling and, 133-134
gradient correlation and, 151
improvement of, 265-276
information and, 240-241, 266-267, 271-272, 274-276
as interpolation, 239, 240, 242
model mismatch and, 248
modularity and, 255
momentum and, 137
noise and, 247-248, 273-274, 277-292
overtraining and, 249-251
PAC learning and, 261-263
performance assessment of, 257-264
physical models and, 276
pruning and, 219, 237, 268
regularization and, 266-267, 281-283
replicated networks and, 272-273
rules of thumb for, 219, 241, 245, 246
size and,241-242, 244-247, 248
training time and, 153
variables and, 253-255
VC dimension and, 261-264
weight decay and, 269-270, 285-287
General position, vectors in, 20-23
Genetic algorithm (GA), 178-179
advantages of, 185, 194, 195
applications of, 191-194
constructive methods and, 217
crossover and, 186, 187, 188
disadvantages of, 185, 194, 195
error surface and, 119
example for, 189-191
fitness scaling and, 189
incompatible genomes and, 193-194
mutation and, 186, 187, 188-189
pruning and, 193, 235
steps in, 186-187
variations of, 187-188
Genetic code, 185, 186
Genetic programming, 193
Genomes, incompatible, 193-194
Goals, 155. See also Objective function
Graceful degradation, 4
Gradation, in hyperplane geometry, 18
Gradient-based methods, 158
first-order, 163-169
second-order, 169-175, 182
Gradient correlation, 150-151
Gradient descent, 57-63
best-step steepest, 165-166
classical optimization and, 155, 163-169, 179
conjugate, 166-169, 179, 182, 252
constructive methods and, 212
convergence rate of, 164, 295-297
jitter and, 280
Gradient descent (cont.)
momentum and, 85-87
Gradient reuse, 150

Top of current section
Previous Section Next Section
Books24x7.com, Inc. © 1999-2001  –  Feedback