Page 108 - DCAP506_ARTIFICIAL_INTELLIGENCE

Artificial Intelligence

Notes Therefore to find if we scrutinize medical evidence to analyze an illness. We must know all the
preceding probabilities to locate symptom and also the probability of having an illness
depending on certain symptoms being observed.

!
Caution The set of all hypotheses must be mutually exclusive and comprehensive.

Did u know? Bayesian statistics occurs at the heart of many statistical reasoning systems.
How is Bayes theorem exploited?

 The key is to invent problem properly:
P(A|B) specifies the probability of A specified only B’s evidence. If there is other relevant
evidence then it must also be taken into account.

Herein occurs a problem:
 All events must be mutually exclusive. However in actual world problems events are not
normally unrelated.

Example: In detecting measles, the indications of spots and a fever are associated. This
signifies that computing the conditional probabilities gets multifaceted.
Usually if a prior evidence, p and some new inspection, N then computing

P(p|N H)
P(H|N p) P(H|N) 1

1
P(p|N)
increases exponentially for huge sets of p
 All events must be exhaustive. This signifies that to work out all probabilities the set of
possible events must be closed.

!

Caution If new information occurs the set must be formed afresh and all probabilities
recalculated.
So Simple Bayes rule-based systems are not appropriate for uncertain reasoning.

 Knowledge acquisition is very rigid.
 Too many probabilities required — too large a storage space.
 Calculation time is too large.
 Updating new information is hard and time consuming.
 Exceptions such as “none of the above” cannot be represented.

 Humans are not very good probability estimators.
However, Bayesian statistics still offer the core to reasoning in most of the uncertain reasoning
systems with appropriate enhancement to conquer the above problems.

102 LOVELY PROFESSIONAL UNIVERSITY