23.02.2015 Views

Machine Learning - DISCo

Machine Learning - DISCo

Machine Learning - DISCo

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Hypotheses, estimation of accuracy<br />

(continued)<br />

bias and variance in estimate, 129,<br />

151, 152<br />

errors in, 129-131, 151<br />

evaluation of, 128-129<br />

justification of, in inductive vs. analytical<br />

learning, 334-336<br />

representations of, 23<br />

testing of, 144-145<br />

Hypothesis space, 14-15<br />

bias in, 40-42, 46, 129<br />

finite, sample complexity for, 207-214,<br />

225<br />

infinite, sample complexity for, 214-220<br />

VC dimension of, 214-217<br />

Hypothesis space search<br />

by BACKPROPAGATION algorithm, 97,<br />

106, 122-123<br />

comparison with decision tree<br />

learning, 106<br />

comparison with KBANN and<br />

TANGENTPROP algorithms, 350-35 1<br />

by CANDIDATE-ELIMINATION algorithm,<br />

64<br />

in concept learning, 23-25, 46-47<br />

constraints on, 302-303<br />

by FIND-S algorithm, 27-28<br />

by FOIL algorithm, 286-287, 357-361<br />

by genetic algorithms, 250, 259<br />

by gradient descent, 90-91<br />

by ID3 algorithm, 60-62,64, 76<br />

by KBANN algorithm, 346<br />

by learning algorithms, 24<br />

by LEARN-ONE-RULE, 277<br />

in machine learning, 14-15, 18<br />

use of prior knowledge, 339-340, 362<br />

ID3 algorithm, 55-64,77<br />

backtracking and, 62<br />

CANDIDATE-ELIMINATION algorithm,<br />

comparison with, 61-62<br />

choice of attributes in, 280-281<br />

choice of decision tree, 63<br />

cost-sensitive measures, 75-76<br />

extensions to, 77. See also C4.5<br />

algorithm<br />

inductive bias of, 63-64, 76<br />

LEARN-ONE-RULE, search comparison<br />

with, 277<br />

limitations of, 61-62<br />

overfitting in, 67-68<br />

search of hypothesis space, 60-62, 64,<br />

76<br />

sequential covering algorithms,<br />

comparison with, 280-281<br />

specialized for concept learning, 56<br />

use of information gain in, 58-60<br />

ID5R algorithm, comparison with GABIL,<br />

258<br />

ILP. See Inductive logic programming<br />

Image encoding in face recognition, 114<br />

Imperfect domain theory:<br />

in EBNN algorithm, 356<br />

in explanation-based learning, 330<br />

in FOCL algorithm, 360<br />

in KBANN algorithm, 344-345<br />

Incremental explanation methods, 328<br />

Incremental gradient descent. See<br />

Stochastic gradient descent<br />

INCREMENTAL VERSION SPACE MERGING<br />

algorithm, 47<br />

Inductive-analytical learning, 334-363<br />

advantages of, 362<br />

explanation-based learning and, 330<br />

learning problem, 337-338<br />

prior knowledge methods to alter search,<br />

339-340,362<br />

properties of ideal systems, 337<br />

weighting of components in EBNN<br />

algorithm, 351-352,355<br />

weighting prior knowledge in, 338<br />

Inductive bias, 39-45, 137-138. See also<br />

Occam's razor; Preference bias;<br />

Restriction bias<br />

of BACKPROPAGATION algorithm, 106<br />

bias-free learning, 40-42<br />

of CANDIDATE-ELIMINATION algorithm,<br />

43-46, 63-64<br />

in decision tree learning, 63-66<br />

definition of, 43<br />

in explanation-based learning, 322-323<br />

of FIND-S algorithm, 45<br />

of ID3 algorithm, 63-64,76<br />

of inductive learning algorithms, 42-46<br />

of k-NEAREST NEIGHBOR algorithm, 234

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!