# Sample Questions for AI Exam 2

#### May still change

 ``` ___ What is planning? Why is planning useful? What is Partial Order Planning? Give an example of a description of an Operator. How can you use Backward Chaining in planning? What is meant by Goal Interaction? Why is it a problem? Use this example in your explanation: initial state: On(A, C) & On(B, Table) goal state: On(A, B) & On(B, C) What is the Threatens link used for? What is the Before link used for? What is Hierarchical Planning, and how does it differ from Linear Planning? What does it mean to use a Least Commitment approach? Why is it useful? Why was Situation Logic invented? i.e., why not use straight First Order Predicate Logic for planning? Give an example of a Frame Axiom (In English, not logic). How do you actually produce a plan when planning using Situation Logic? ___ Does Winston's ARCH system do Induction? Explain why or why not. Why does it use Near Misses as opposed to unrestricted negative examples? Describe a possible change a near miss can cause to the evolving model. Do the examples given matter? Does their order? ___ Describe the usual sequence of steps taken during Case-Base Reasoning (CBR). What kind of simple adaptation step is usually assumed? Give an example. Could CBR be used in the adaptation step. Describe how it might work. Contrast the feasibility of case-based menu planning with case-based soccer. ___ What kind of examples have to be provided when using the Version Space approach (i.e., multiple models) to learning? What form do they have? What kind of example specialize general descriptions? If G = { ( ? ? ? ) } , explain why specialization is necessary when an example of that kind is given. Can the Version Space approach handle noisy data? Explain why or why not. Explain whether this method can be used for learning an Arch model. ___ Is "Learning by Building Identification Trees" a supervised or unsupervised type of learning? Is it inductive or deductive? What is an Identification Tree? (IDT) Contrast a "training set" with a "test set". What type of training set is used when learning IDTs? Explain how the concept of Average Disorder helps build a tree. How do you know when to stop building the IDT? How do you get rules as the final result? Can the IDT learning approach handle noisy data? Explain why or why not. ___ Explain what mutations and crossovers are in a Genetic Algorithm. Describe the basic GA process for finding solutions. What is fitness? What are the benefits of using the Rank-Space Method for determining fitness? Koza's approach to Genetic Programming used a representation for programs that enabled crossovers between programs. Describe the representation. ___ In the simple form of the Object Recognition approach using templates described by Winston, what are two of the simplifying assumptions? Given two stored images (I1 and I2) of an object and an unknown image (I0) explain what you need to have in order to find Alpha (A) and Beta (B) in the equation XIu = A.XI1 + B.XI2 Once A and A are found, how can the equation be used to recognize an unknown? ___ Why doesn't edge finding using the 2nd derivative work very well with a "real" image? How would you make it work better? (i.e., to produce the sombrero operator). What is the significance of a zero-crossing? How can a sombrero be made wider or narrower, and what effect does that have when convolved with an image? Why would you want to find disparity in two stereo images? Briefly describe the process for finding corresponding features in two stereo images. Why would you use sombrero operators with two widths in that process? What is the key property of a Lambertian surface? What two measures does an FG reflectance map relate? How could you generate a Synthetic Image using a reflectance map? Why do you need a relaxation procedure in order to determine surface direction from surface shading? ___ Why is the fact that people can spot incorrect examples important? Give the names and examples of three types of "word groups". What are Syntax, Semantics and Pragmatics? What is surface structure, as opposed to deep structure? Give a simple Context Free Grammar for an English (or American) sentence (include only non-terminals, and not more than 6 rules), and an example of what it can parse. Compare the X-bar structure for "the library in the city" and "all return their books". Contrast language "Competence" against "Performance". ___ What additional capabilities does an Augmented Transition Network add to a Context Free Grammar? Explain the use of one of these capabilities. Draw a subnet for the grammar rule: NP --> Det Adj* Noun PP* What is a Semantic Grammar? ___ ```