Final Exam Sample Questions

The materials of the programming assignments are testable.

Also see the sample questions from Midterm 1 and Midterm 2.

The contents of Reading 1, Reading 2, and Reading 3 are testable.

The contents of Exercise 1, Exercise 2, and Exercise 3 are good possible question types as well.

1. What are the components of a production system?

2. Describe the Rete algorithm? What is it used for? Why is it used?

3. What is the STRIPS notation used in planning systems? Give an example STRIPS-type operator.

4. In Brooks' subsumption architecture, explain the concept of a layer and a module. Explain how one layer can connect to another. Can one layer influence another? How? In what way(s)?

5. Compare and contrast Brooks' subsumption architecture with the general assumptions of the Physical Symbol System Hypothesis with respect to sensing/perception, cognition/thinking/processing, and action/behavior.

6. Prove the conditionalized version of Bayes' rule: P(A | B ^ C) = P(B | A^C) P(A | C) / P(B | C)

7. Explain why: P(X) = P(X ^ ~ Y) + P(X ^ Y), where ~ is negation.

8. Sensors in robotic or biological systems generally have noise. That is, there will be some error associated with the information obtained from the sensors. Logical systems such as resolution theorem provers assume a complete and consistent set of wffs in their KB's. There is an apparent conflict between these two issues. On the one hand, sensors have noise. On the other hand, logical systems require complete and consistent formulas. Suppose you are designing an AI system that operates in the real world (e.g., a robotic system):

(a) State and describe two problems that would likely arise from this apparent conflict.

(b) Suggest ways that we might use to get around these problems.

9. Suppose you are representing a feature, S, size, with feature values: huge, large, medium, small, tiny.

(a) How many bits of information will it take for you to represent a single item's category?

(b) Say that we know some frequency information about the size category. Suppose P(S=huge) = .2, P(S=large) = .3, P(S= medium) = .4, P(S =small) = .04, P(S=tiny)= 0.06. Calculate the average information required to represent an item's category, given this frequency information, and given that we are representing many items.

10. In computer vision systems, there can be influences both from sensors and models of the world (e.g., "expectation"). Consider the following events with a computer vision system that is detecting the presence or absence of an object.. Let O be the (boolean) event that an object is present. Let E be the (boolean) event that we are expecting an object to be present. Let S be the (boolean) event that our camera sensor indicates the object is present. Let Sf be the (boolean) event that our camera is faulty. Let D be the (boolean) event that the vision system detects the object. The presence of an object (O) influences the vision sensor (S). The system's "expectation" (model) influences whether the system detects the object (D).

(a) Draw the Bayesian belief network corresponding to this situation.

(b) Given the following prior and conditional probabilities, compute the probability that we detect an object given that the camera is not faulty and we expect an object. P(Sf) = 0.01; P(O) = 0.7; P(E) = 0.5