PSU CS410/510SS: Search and Scheduling
Lecture 3
Systematic and Local Search Techniques

Administrative

• Library reserve
• 3 vs. 4 credits
• HW1 + email
• Extra credit
• HW2
• Korf: bad bug at 4.5 (missing material)

Lecture 3

• Korf: ``Best-First'' Search
  • Pathfinding problems vs. CSPs
  • Important idea: Admissible heuristic gives lower bound on distance to solution.
  • Trees vs. graphs
  • Open and Closed lists
  • Dijkstra's Algorithm
    • Algorithm
    • Complexity (O(n<sup>2</sup>)?)
  • Pure Heuristic Search
    • Very fast
    • Does not work
  • A*
    • f(n) = g(n) + h(n)
    • h consistent
    • consistency = local consistency = monotonicity => admissibility
    • achieving consistent from admissible
    • Danger: may regenerate open node, so take min
    • Danger: break ties in f toward larger g
    • Analysis
      • Worst-case
      • Analytic expected case
      • Korf's expected case (3.7.5)
    • A* does not work! (memory)
  • IDA*
    • Like DFID with f = g + h
    • Increment by minimum unselected
    • Problems
      • Cycles in graphs (to close or not?)
      • Constant time factor d
  • DFBnB
    • For CSPs
    • Bound exploration by best soln so far
    • ``Anytime'' property: computes successive approximations
  • RBFS (not required)

• Systematic CSP Search
  • Variable ordering heuristics
  • Value ordering heuristics
  • Node selection
    • LDS
    • DDS