Review session covering text book pages 145150 and 155157

1
Review session covering text book pages 145-150
and 155-157

• A f(n)g(n)h(n)f(n), g(n), h(n) are the
  true path cost, whereas f(n), g(n), h(n) are
  estimations of the true cost.
• Heuristics that find the shorts path to a goal
  are called admissible. How do we choose and
  compare admissible heuristics?
• Compare the counting tiles out of place vs. the
  tile Manhattan distances to correct position
  heuristic for the 8-puzzle. Which is better and
  why?
• Monotonicityalgorithm consistently finds
  shortest path for pairs of states it encounters
  during the search If a heuristic has this
  property, if the same state is encountered a
  second time, the second path is never optimal
  e.g. the algorithm can terminate when it
  encounters the goal state the first time note
  that A terminates (see page 134) when it tries
  to expand the goal state A expands states in
  non-decreasing f-value order and for each
  expanded node g(n)g(n).
• Does admissibility imply monotonicity?
  Interesting theoretical question!

2
Review session covering text book pages 145-150
and 155-157

• The set of states expanded is smaller for a more
  informed heuristic---see figure 4.18!
• Games Search deep with simple heuristics vs.
  Search not so deep with sophisticated heuristics.
• Minimax requires a 2-path analysis of the search
  spaceexplain!
• What state generation order is most beneficiary
  for alpha-beta?

3
Relevant MaterialExam1 Tu., Feb. 26, 2008

• Remark Exam is open-notes/open-books
• Relevant Textbook pages 1-4, 20-31, 79-108,
  123-128, 132(Section 4.2)-161, 385-386, 507-534.
• Relevant Transparencies Introduction to AI Dr.
  Eick's Introduction to AI Heuristic Search
  Luger Chapter 3a, Luger Chapter 3b, Eick Search1,
  Eick Search1b(new), Eick Search2, Luger Chapter
  4a, Luger Chapter 4b, Luger Chapter 4c, Eick
  Search3. Machine Learning Tom Mitchell's
  Introduction to Machine Learning, EC(EvoNet EC
  Introduction, Using EC to Solve TSP, Brief
  Introduction to Evolutionary Machine Learning,
  Luger Chapter 12)
• Other Material knowledge Nilssons backtracking
  algorithm, Mitchells the discipline of machine
  learning.