; k5
Documents
Resources
Learning Center
Upload
Plans & pricing Sign in
Sign Out
Your Federal Quarterly Tax Payments are due April 15th Get Help Now >>

k5

VIEWS: 0 PAGES: 23

  • pg 1
									Chapter 5: Representing Cases

•   To represent a case, we need to consider:
    • What are the parts of a case?
    • What kinds of knowledge are included?
    • How do we encode this knowledge?
    • How do we decide how much knowledge can fit in a case?
•   The major components of a case are:
    • The problem or situation description
    • The solution or situation reaction
    • The outcome of applying the solution or acting on the interpretation
•   Not all CBR systems have all components, but what you include in a case
    directly impacts what you can use the case for




                  CS 682, AI:Case-Based Reasoning, Prof. Cindy Marling   1
Representing the Problem

•   In representing the problem, you need enough detail to be able to
    determine if an old case and a new one are similar
•   Aspects that may be included are:
    • The goals to be achieved
    • Constraints on achieving the goals
    • Features of the problem situation and the relationships between its
         parts
•   The following examples are loosely patterned on MEDIATOR, CHEF
    and PROTOS




                   CS 682, AI:Case-Based Reasoning, Prof. Cindy Marling     2
MEDIATOR Example




           CS 682, AI:Case-Based Reasoning, Prof. Cindy Marling   3
CHEF Example




           CS 682, AI:Case-Based Reasoning, Prof. Cindy Marling   4
PROTOS Example




           CS 682, AI:Case-Based Reasoning, Prof. Cindy Marling   5
Representing the Problem or Situation

•   You always need to represent the problem or situation
•   What you include in it is domain dependent
•   General guidelines are:
    • Include all information that was explicitly taken into account in
       solving the problem
    • Include all information that is normally used by people when
       describing this kind of problem




                   CS 682, AI:Case-Based Reasoning, Prof. Cindy Marling   6
Representing the Solution

•   You also need to represent the solution
•   Depending on the problem domain, the solution might be a plan, a
    classification, or any number of other things
•   You at least need the solution itself. Some CBR systems also include:
    • The set of reasoning steps used to solve the problem
    • The set of justifications for decisions that were made in solving the
        problem
    • Acceptable solutions that were not chosen and the reasoning and
        justifications for them
    • Unacceptable solutions that were ruled out and the reasoning and
        justifications for ruling them out
    • Expectations of what will result upon deployment of the solution



                   CS 682, AI:Case-Based Reasoning, Prof. Cindy Marling       7
Representing Outcomes

•   Besides problems and solutions, cases may also include outcomes
•   Outcomes may include feedback from trying the solution out in the
    world
•   Cases may include:
    • The outcome itself
    • Whether or not the outcome fulfilled or violated expectations
    • Whether the outcome was a success or a failure
    • If the outcome was not successful, you may store:
        • An explanation of the expectation violation or failure
        • A repair strategy
        • What could have been done to avoid the problem
        • A pointer to the next attempted solution, that is, the result of
            applying the repair

                   CS 682, AI:Case-Based Reasoning, Prof. Cindy Marling      8
Example from MEDIATOR




           CS 682, AI:Case-Based Reasoning, Prof. Cindy Marling   9
MEDIATOR Example, continued




            CS 682, AI:Case-Based Reasoning, Prof. Cindy Marling   10
MEDIATOR Example, continued




           CS 682, AI:Case-Based Reasoning, Prof. Cindy Marling   11
Case Representation in CASEY

•   CASEY stores problems and solutions, but does not store outcomes
    • It doesn’t receive any feedback, so it can’t repair mistakes
•   The following examples show:
    • CASEY’s problem description, a list of data about a patient
    • CASEY’s solution, a causal network
    • CASEY’s justification of its solution, a comparison of the new and
        the old patient cases




                  CS 682, AI:Case-Based Reasoning, Prof. Cindy Marling   12
CASEY’s Problem Description




             CS 682, AI:Case-Based Reasoning, Prof. Cindy Marling   13
A Solution in CASEY




            CS 682, AI:Case-Based Reasoning, Prof. Cindy Marling   14
Solution Justification in CASEY




              CS 682, AI:Case-Based Reasoning, Prof. Cindy Marling   15
Solution Representation in CHEF

•   Because a solution is a plan in CHEF, its representation differs from
    those seen in other systems




                   CS 682, AI:Case-Based Reasoning, Prof. Cindy Marling     16
Part of a Case from CHEF




             CS 682, AI:Case-Based Reasoning, Prof. Cindy Marling   17
Case Representation in JULIA

•   JULIA’s case representation allows it to reason in its own domain
•   JULIA represents the constraints that must be satisfied as part of the
    problem description
•   JULIA breaks up each case into components so that it can use parts of
    multiple past cases in forming new designs
•   JULIA records the reasoning that indicates adaptation may be needed




                   CS 682, AI:Case-Based Reasoning, Prof. Cindy Marling      18
A Case in JULIA (continued on the next slide)




               CS 682, AI:Case-Based Reasoning, Prof. Cindy Marling   19
CS 682, AI:Case-Based Reasoning, Prof. Cindy Marling   20
Simpler Representations

•   Commercial applications typically use simpler representations and
    reasoning
•   If a problem is simple, like choosing a gift or a movie to see, or if a
    human being interacts with the system to provide some natural
    intelligence, a simple representation may suffice
•   The simplest representation is a form filling representation, that makes
    each case look like a database record
•   Battle Planner uses a simple representation in which a battle is
    represented by a list of features
    • Battle planning is not simple, but a person does most of the
         reasoning
    • The system retrieves the most similar historical battle to the battle
         plan described by the user
    • If the past battle was lost, the user manually adapts their own plan

                   CS 682, AI:Case-Based Reasoning, Prof. Cindy Marling    21
A Case in Battle Planner (continued on the next slide)




                 CS 682, AI:Case-Based Reasoning, Prof. Cindy Marling   22
CS 682, AI:Case-Based Reasoning, Prof. Cindy Marling   23

								
To top