Final Exam Study Guide The Final Exam is Thursday, Dec 13 from 9

Final Exam Study Guide
The Final Exam is Thursday, Dec 13 from 9:00 AM – 12:00 Noon (usual room).
The final exam is comprehensive – everything that you were responsible for on the
midterm exams, you will be responsible for on the final exam. You will also be
responsible for material since the latter midterm exam. The midterms give you a
very good idea of the form and content of many questions, though because the final
exam needs to be graded very quickly, you may have a larger portion of multiplechoice questions. My multiple choice questions typical require you to do some
problem solving and choose all valid answers (e.g., given different tie-breaking
policies, what are sequences of states that might be visited by best first search on
the given graph).
In terms of content, here are some guarantees. You definitely will see questions
a) asking you to “hand trace” various search strategies (and a 100% guarantee
that one of them will be depth first branch and bound using f (i.e., actual cost
plus h),
b) asking you to illustrate the generalized arc consistency algorithm,
c) asking you to prove a propositional statement using specified strategies (and
certainly at least one using resolution refutation),
d) asking you to express joint and conditional probabilities given a Bayesian
Belief Network),
e) asking you to hand trace game-tree search with alpha-beta pruning.
LISP. If I ask you any LISP questions, then will be small variations on LISP questions
that I have asked you on previous exams.
The following are the sections from the Poole and Mackworth for which you are
responsible on the final exam. Under some of these, I list a “heads up” on some
specific material that might come up in the final exam (in addition to that already
mentioned). In general, make sure that you look over exercises that you were
required (to turn in) or those that you were to do but not turn in.
Chapters 1, 2 – know terms, concepts, and examples – I won’t draw substantially
from these chapters, but I might draw one or two questions (e.g., the anytime search
question on the first exam corresponded very closely to example 1.9 of the text)
Chapter 3
Heads up: in addition to all methods covered on previous exams you’ll be
responsible for Dynamic Programming; don’t forget AI assignment exercises, both
required and “optional”. Heads up: Anytime search is not a search strategy per se,
but a capability or criteria for assessing search strategies; an anytime search doesn’t
stop when a solution is found – an anytime search keeps looking, and at any point in
time, is able to report the best solution found so far. Consider heuristic-DFS Branch-
and-Bound search as an anytime search algorithm. Contrast it with A* as an anytime
search algorithm.
Chapter 4 through 4.8, 4.11 – 4.13
Heads up: in addition to GAC (AI assignment ex 4.3), be able to do variable
elimination (see posted problem key – variable elimination may be asked as a low
weight problem), and understand how constraint reasoning can be performed with
various state space search strategies.
Heads up: Consider local search (section 4.8) on solving SAT (satisfiability)
problems: given a statement in propositional logic, say in conjunctive normal form,
decide whether there is at least one assignment of propositions to truth values that
makes the whole statement true (see Chapter 5 and propositional logic lecture
slides). Formulate a local search with random restart solution to the problem of
finding an assignment of truth values to propositions that satisfied a propositional
statement in conjunctive normal form.
Chapter 5 through 5.2 (plus propositional logic lecture slides )
Heads up: know other bottom-up and top-down propositional proof strategies, and
very particularly resolution theorem proving
Chapter 6 through 6.3, 6.4.1
Heads up: in addition to Bayesian net problems like those in the midterm (and AI
exercise 6.6), also be able to construct a Bayesian network (e.g., recall the in-class
exercise of creating a network of CS courses).
Chapter 7 through 7.3.1
Heads up: the focus will be on decision tree induction; understand the problem of
over-fitting (see in-class exercise) and its implications for decision tree induction;
know how to estimate probabilities as proportions revealed in data
Chapter 8 through 8.3
Heads up: After project 3, I’m guessing you will understand forward planning (and I
may ask that you demonstrate it), but be able to hand trace a regression planner in
propositional STRIPS representations (Example 8.9), and first order STRIPS
representations. There is a very good chance that I will ask you to find the
preconditions and effects (to include in terms of add and delete lists) for a short
macro operator of 3-4 operators, in both propositional form (e.g., exercise 8.7) and
first-order form (lecture slides). There is a good chance that I will ask you
Chapter 10 through 10.3
Heads up: Be able to hand simulate min-max search with alpha beta pruning (see
slides) – you will definitely be tested on this; you will not be tested on the
probabilistic version of mini-max search
Chapter 11 (11.1 only)
Given two examples (feature vectors) be able to compute the sum-of-square errors;
know example 11.2
Chapter 12 through 12.3
Questions related to first order planning, as previously outlined. Be able to identify
whether operator conditions are satisfied by a particular world state or not.
Chapter 14.1
Heads up: be able to hand simulate forward and regression planning in a first order
STRIPS representation (slides will be more helpful than reading); be able to obtain
the PRE, ADD and DEL lists for a macro operator in first order representation. Be
able to show the children or grandchildren of a state (or list of conditions) using a
progression (or regression) planner.
Other
I want to add additional questions to the final exam, intended to assess (a) whether
you can reason about tradeoffs in algorithm design from consideration of math
foundations, algorithmic principles and theory; the most obvious place to look for
questions concerned with reasoning about tradeoffs in system design would be
projects 1 and 2. I also want to ask you questions that cause you to reflect on your
professional development. We’ll discuss question possibilities on the Oak Discussion
Board, converging on some guaranteed questions by next Tuesday.