Lec 5-6 Flashcards
What is an expert system?
A computer system that emulates decision making process of domain expert.
- This requires encoding a great deal of domain-specific knowledge
- Reasoning often encoded using vast knowledge base of rules.
-Process often augmented by heuristics.
What are the main takeaways of expert systems?
Expert systems are specialized systems aiding/emulating decision making. They encode expert knowledge, often in rule-based form. Together with inference engine, knowledge is used to draw conclusions. Search may be sped up using heuristics.
What are heuristics and what’s their use?
Criteria, methods, or principles for deciding which among several alternative courses of action promises to be the most effective in order to achieve some goal.
Rules of thumb, not always applicable or optimal, quickly finding something “good enough”, compromise:
Between simplicity, efficiency, accuracy and completeness
Symbolic AI in 1980’s: Commercialization of Expert Systems
Proved a way to get AI from labs to business
Connectionist AI: Resurgence in 1980’s
With increasing compute power, models start to be applicable to “real” problems
1980’s Second Boom:
Return of funding, investment in fifth generation computer systems that were to…
- use super large scale integrated chips
- have AI
- ability to recognise images and graphs
- solve complex problems
- work with natural language
What are limits of the expert system?
- Classical formal logic not suited for all problems
- Performance issues on large knowledge base
- Expert elicitation - translating expert knowledge to a computer is very yhard
- Many experts don’t reason in if-then rules
What caused the second AI winter? 1987-1993
The over-hyped but disappointing developments. Funds were not extended, researchers embarrassed or less interested.
What can cause uncertainty?
Inherent randomness in outcomes, unknowns/uncertainties in domain description, existence of many special cases
How can uncertainty be represented?
Probability theory, and extensions and generalisations of this. Fuzzy sets and logic or non-monotonic logics
What is a sample space?
A non empty set, representing the ‘atomic things that can happen’.
What is an event?
Something that can happen, therefore a subset of the sample space.
A sigma algebra over the sample space is a set of events such that..
It contains the sample space.
It is closed under complements.
It is closed under countable unions. — therefore also closed under countable intersections
The tuple of the (SampleSpace, SigmaAlgebra) is a measurable space.
Sigma Algebra can be considered a set-theoretic complement, union, and intersection
If we have a dice, the sample space is {1,2,3,4,5,6}. Then the event
(#Eyes is odd AND #eyes >= 2) = (#Eyes is odd) AND (#Eyes >=2) = {3,5}
What is the triple (S,SA,P) called?
Probability space
