Page 173 - DCAP506_ARTIFICIAL_INTELLIGENCE

Unit 12: Learning

16. ......................... starts by doing trait withdrawal on data items and compute the values of Notes
the selected feature set.

12.10 Learning by Correcting Mistakes

When learning new things, there is a likelihood that the learning system may make errors.
Similar to human beings, learning system can correct itself by determining reasons for its
breakdown, isolate it, elucidate how the particular supposition causes failure, and alters its
knowledge base.

Example: When playing chess a learning system may make an incorrect move and
finishes up with failure. Now the learning system considers the reasons for the breakdown and
corrects its knowledge base. So when it plays again it will not replicate the similar mistake.
In his work, Active Learning with Multiple Views, Ion Muslea has utilized this technique to
label the data. He produces a technique called Co-EMT which is a amalgamation of two techniques:
Co-testing and Co-EM. The Co-testing technique communicates with the user to label the data.
If it does any mistake in labeling, it learns from the mistakes and enhances. After learning, the
system labels the unlabeled data removed from a source proficiently.

Did u know? The labeled data comprises what is known as knowledge.

Self Assessment

Fill in the blank:
17. When learning new things, there is a likelihood that the learning system may make
......................... .

12.11 Learning by Recording Cases

A program that learns by recording cases usually use constancy heuristic. As per constancy
heuristic, a property of something can be estimated by locating the most similar cases from a
specified set of cases.

Example: A computer is given the images of different types of insects, birds, and animals.
If the computer is asked to recognize a living thing which is not in the recorded list, it will
contrast the specified image with already recorded ones, and will at least tell whether the
specified image is insect, bird or animal.

Learning by recoding cases method is chiefly used in natural language learning tasks.
For the period of the training phase, a set of cases that illustrate vagueness resolution episodes
for a specific problem in text analysis is gathered. Every case includes a set of traits or attribute-
value pairs that instruct the context in which the vagueness was encountered.
Furthermore, every case is annotated with solution traits that enlighten how the ambiguity was
resolved in the present example. The cases which are formed are then accumulated in a case base.
Once the training is over, the system can make use of the case base to resolve ambiguities in new
sentences. This manner, the system obtains the linguistic knowledge.

LOVELY PROFESSIONAL UNIVERSITY 167