1. Problem
Representation and
Problem-solving
Strategies

2. Cognitive science（認知科學） -- A study devoted to
Problem Solving – performance output of thinking creatures and the term was introduced by mathematicians
Decision making = problem solving
Humans follow a fairly standard process when they solve problems (see Figure)
Cognitive science（認知科學）
-- A study devoted to
learn how we learn, and know that our strengths and weaknesses are revealed;
understand our minds, which can lead to improved ways of learning and applying our intelligence to real-world problems
Newell-Simon Model (see figure)

3. Step 1 Step 2 Step 3 Step 4 Step 5 Step 6 Implementation
Problem Identification
And Definition
Identifying Evaluation
Criteria
Generation of
Alternatives
Search for Solution
And Evaluation
Choice and
Recommendation
Implementation
Step 1
Step 2
Step 3
Step 4
Step 5
Step 6

4. Short-term (Working) Memory External Memory: Paper, Chalkboard
Perceptual Subsystem
Input Stimulus
Sensors
Buffer Memories
Cognitive Subsystem
Long-term Memory
Short-term (Working) Memory
External Memory: Paper, Chalkboard
Elementary Processor
Interpreter
Motor Subsystem
Human Muscles
Buffer Memories
Output Responses
Newell-Simon Model of Human Information Processing. (Adapted from A. Newell and H. A. Simon, Human Problem Solving, Prentice-Hall, 1972)

6. Problem Solving in AI
Primarily with the search and evaluation steps of the problem solving process
Search Approaches
Optimization
attempts to find best possible solution by using mathematical formulations that model specific situations
mathematical approach that using a one step or an algorithm; e.g. Gradient descent method, Conjugate gradient method, BFGS method, ….
Bio-computing approach; e.g. Genetic algorithm
Blind search
Exhaustive (complete) search
Partial search
Use of heuristics

8. Search Process Test Solution Search approaches Optimization
(Analytical)
Blind
Search
Heuristics
Complete
Enumeration
Partial
Search
Generate
Improved solutions or get the best solution directly
All possible solution are checked
Check only some alternatives systematically drop inferior solutions
Only promising solutions are considered
Search
Process
Stop when no improvement is possible
Comparisons stop when all alternatives are checked
Comparisons
Stop when solution is good enough
Test
Optimal (best)
Optimal (best)
Best among alternatives checked
Good enough
Solution

9. Blind search methods
--Arbitrarily to search a state space and can be classified as
Exhaustive search
Partial search
Breadth-first method（先廣後深）
Depth-first method（先深後廣）
Blind search
Exhaustive
Partial
Breadth first
Depth first

10. Search Directions Data-directed (Forward) search –
A search starts from available information (or facts) and tried to draw conclusions regarding the situation or the goal attainment.
Goal-directed (Backward) search –
A search starts from expectations of what the goal is or what is to happen; then it seeks evidence that support those expectations (or hypothesis).

11. Problem representation in AI
(not Knowledge representation)
Three major elements: problem states, a goal, and operators.
States – snapshots of varying conditions in the environment
Goal – the objective to be achieved, a final solution
Operators – procedures used for changing from one state to another.
A control strategy – selects or guide the procedures
The selection of a particular knowledge representation method will greatly effect the type of search and control strategy used.
State-space representation
A state space representations the set of all attainable states for a given problem

12. Relationship between initial state(s), procedures, and goal(s) in the search process
Control strategy
Goal(s)

17. Heuristic methods
To reduce the amount of search for a solution by pruning nonvital or non-promising nodes
The solution is not optimal, so it is termed good enough
Representative approaches
Generate and Test – generate possible candidate solution and devise a test to determine if the solution are indeed good
Hill Climbing – similar to depth-first blind search, however, paths are not selected arbitrary, but in relationship to their proximity to the desired goal. (Example)

18. Best First (A *)
A search approach based on some heuristic evaluation function, users select the next move by searching for the best available solution that you can move to in one step, no matter where it is located on the tree.

19. Use of evaluation function
Comparison of search strategies
Approach
Use of evaluation function
Delete the candidates
Search efficient
Search quality
Depth first No
Bad
Good
Breadth first
Hill climbing
Yes
Best first
Middle

21. Example

22. Depth-first (先深後廣)

23. Breath-first (先廣後深)

24. Hill-climbing (登山捜尋法)

25. Best-first (A* 最佳優先法

30. Level of search
g(n)=
g(n)=0
g(n)=1
g(n)=4
g(n)=2
g(n)=3
g(n)=5