Russell and Norvig, Chapter 3: Solving Problems by Searching

3.1 Problem-Solving Agents [Fig3.1]
- update state
- formulate goal
  - e.g., get gold, get home
- formulate problem
  - actions, states, initial/goal states
- search -> action sequence
- return action

3.2 Formulating Problems
- problem types
  - single state
    - accessible world
    - agent always knows what one state it is in
  - multiple state
    - inaccessible world
    - agent knows which set of states it can be in
  - contingency
    - action depends on what is found to be true
  - exploration
    - no knowledge of states or actions
    - experimentation
- problem definition (multi-state)
  - initial state (set of states)
  - operators (determine set of states reachable)
  - goal test (satisfied by every state in the set)
  - path cost function
- measuring problem-solving performance
  - correctness
  - search cost = time/memory to find solution
  - total cost = path cost + search cost

3.3 Example Problems
- toy problems
  - 8 puzzle
  - 8 queens
  - cryptarithmetic
  - vacuum world
  - missionaries and cannibals
- real world problems
  - route finding
  - touring and travelling salesperson problems
  - VLSI layout
  - robot navigation
  - assembly sequencing

3.4 Searching for Solutions
- search tree
  - select node on frontier (search strategy)
  - if not goal, expand node

3.5 Search Strategies
- breadth-first search
- uniform-cost search
  - select node with least path cost so far
  - costs = 1 --> breadth-first
- depth-first search
- depth-limited search
  - how to choose depth limit
- iterative-deepening search
  - for d = 0 to inf {depth-limit-search(d)}
  - most nodes at last level, so repeating previous levels not bad
- bidirectional search
  - how to reverse operators
  - how to detect a connection
    - remember frontier (expensive)
- performance comparison [Fig3.18]

3.6 Avoiding Repeated States
- do not return to your parent state
- do not return to an ancestor state
- do not return to any previously-visited state
  - b^d memory
  - hash function to speed access

3.7 Constraint Satisfaction Search
- state = assignment to a set of variables
- goal test = constraints on variable values
- depth-first search
  - solution depth = number of variables
  - operators consist of assigning possible values to one variable
  - big branching factor
- backtracking search
  - apply tests after each assignment, rather than after all variables
    assigned
- forward checking
  - as variables are assigned values, removes conflicting values from the
    domains of unassigned variables
  - reduces branching factor
  - backtracking occurs earlier
- arc consistency and constraint propagation
  - continually remove values from domains inconsistent with constraints