Documents
Resources
Learning Center
Upload
Plans & pricing Sign in
Sign Out

Case-Based Reasoning 2

VIEWS: 1 PAGES: 4

									             Case-Based Reasoning                                           Case-Based Reasoning
• case used to store description of past                       • contextualised piece of knowledge which teaches
  experience                                                     the reasoner a lesson
• problem encountered and its proposed                         • some areas of artificial intelligence
  solution                                                          – effort first goes into developing a model of how a
                                                                      thought or decision-making processes works
• may represent a single past case or a
                                                                    – then, that model is generally applied to all problem
  generalisation of several single cases                              situations involving that particular process
• Case-based reasoning is not the first                             – the knowledge in the model will be general knowledge
  artificial intelligence method to combine
  reasoning and learning
October 06             Case -Based Reasoning               1   October 06                Case -Based Reasoning                   2




             Case-Based Reasoning                                           Case-Based Reasoning
• general knowledge                                            • cases represent knowledge at the operational level
     – advantage of economy of storage                         • remembered if the experience is instructive
     – allows to deal with uncertainty using statistical            – whether in the positive or negative sense
       models                                                       – can offer advice for potential future success or failure
     – disadvantage of difficulty when trying to apply         • a model of reasoning that makes use of problem
       general rules to something specific                       solving, understanding, and learning which
                                                                 integrates these three processes within something
     – doesn’t tell us about situations that deviate from
                                                                 called memory
       the norm


October 06             Case -Based Reasoning               3   October 06                Case -Based Reasoning                   4




             Case-Based Reasoning                                           Case-Based Reasoning
• A Aamodt (1989) paper “Towards Expert                        • Delaney, Cunningham, Coyle (2004) paper “An
  Systems that learn from experience”                            Assessment of Case-Based Reasoning for Spam
     – case features categorised as necessary,                   Filtering”
       characteristic, non-characteristic and irrelevant            – The case base can be updated continuously and new
                                                                      training data is immediately available
• K Sycara (1989) “Index Transformation and
  generation for case retrieval”                               • McSherry (2004) paper “Maximally Successful
                                                                 Relaxations of Unsuccessful Queries
     – consider structural features, a functional
       description, causal explanation of behaviour                 – Presents a mixed initiative approach to recovery from
       and qualitative states                                         the retrieval failures that occur when there is no case
                                                                      that satisfies all requirements

October 06             Case -Based Reasoning               5   October 06                Case -Based Reasoning                   6
               Case-Based Reasoning                                                        Case-Based Reasoning
• According to a 1989 panel on CBR, some                                         – if dependency structure and causal annotation
  important questions to ask about case                                            appear in case representation at all, when
  representation include                                                           should the relevant information be acquired?
     – to what extent should cases be generalised as they are                            • at storage time, time of modification, use
       stored?
                                                                            • CBR has been implemented in many forms
     – What argument is there for maintaining the distinctness
       of cases that appear similar?                                             – decision support systems
     – Are cases monolithic structures that are applied                          – groups of co-operating application processes
       individually or are they loosely connected sets of events                   sharing information in OO database
       that are reconstructed at retrieval time?                                 – can be used to support people in tasks
October 06                    Case -Based Reasoning                     7   October 06                    Case -Based Reasoning                 8




               Case-Based Reasoning                                                        Case-Based Reasoning
• PERSUADER                                                                 • ECUE
     – knowledge based system used to model the                                  – E-mail Classification Using Examples
       dynamics of negotiation                                                   – A lazy learning system using CBR
     – input is goals of each side and the dispute                                       • Lazy Learning – the decision of how to generalise
       context                                                                             beyond training data is deferred until each new
                                                                                           unseen instance is considered.
     – CBR generates initial settlement, persuasive
       arguments and improving rejected proposal
             • responses of negotiating parties will lead to further
               transformation of solution

October 06                    Case -Based Reasoning                     9   October 06                    Case -Based Reasoning                10




                      Expert Systems                                                              Expert Systems
• developed as specialised problem solvers                                  • have their knowledge encoded and maintained
  that emphasised the use of knowledge                                        separately from computer program which uses the
                                                                              knowledge
     – medical diagnosis
                                                                            • capable of explaining how a particular conclusion
     – mineral prospecting
                                                                              is reached
• designed to reason through knowledge                                      • use symbolic representation for knowledge
     – solve problems using methods that humans use                         • perform inference through symbolic computations
• use heuristic knowledge rather than number                                     – closely resemble manipulations of natural language
  to control the process of problem solving
October 06                    Case -Based Reasoning                    11   October 06                    Case -Based Reasoning                12
                        Expert Systems                                                                              Expert Systems
• knowledge engineer extracts knowledge                                                          – inference engine
  from expert and places it in the knowledge                                                     – working memory
                                                                                                         • store user’s input, some rules and other pertinent facts
  base
                                                                                                 – I/O interface
• knowledge engineer develops the inference
                                                                                                 – explanation module
  engine                                                                                                 • true expert systems are capable of explaining what they are
     – sorts through the knowledge in organised                                                            doing at any point in the process
       manner                                                                                            • how it arrives at solution

• components in expert system                                                                    – editor
                                                                                                         • add or change rules and knowledge base
     – knowledge base
October 06                        Case -Based Reasoning                                13   October 06                       Case -Based Reasoning                       14




                        Expert Systems                                                                              Expert Systems
     – learning module                                                                                   • methods developed to cope with this include
             • some expert systems can include learning module                                             reasoning by elimination, abstraction, multiple lines
             • not common                                                                                  of reasoning and least commitment principle
• each rule in the knowledge base represents                                                     – elimination
  small part of knowledge in the domain of                                                               • discard rules that do not lead to solution or lead to
                                                                                                           solution of low plausibility
  expertise
                                                                                                 – abstraction
• weaknesses
                                                                                                         • break the problem into sub problems, sub sub
     – don’t perform well with large number of rules                                                       problems etc
       or large search spaces                                                                            • then solve lowest level and work upwards

October 06                        Case -Based Reasoning                                15   October 06                       Case -Based Reasoning                       16




                        Expert Systems                                                                              Expert Systems
             • uses guessing when impossible to determine which is the best                      – ES tries to take account of quality of
               rule to select at given point
                 – if it guesses wrong, it must be able to recognise this and try to
                                                                                                   knowledge
                   recover                                                                               • can use probability theory or incidence calculus to
                                                                                                           deal with uncertainty
     – multiple lines of reasoning
             • view the problem from different perspectives                                      – often make use of commercially available
             • try to solve and compare solutions                                                  Expert System Shells of which criticisms
• weaknesses                                                                                       include
                                                                                                         • not always capable of versatile searching
     – do not present uncertainties very well
                                                                                                         • some do not handle uncertainty well
     – knowledge may be incomplete, unreliable, imprecise or                                             • some crunch through large collections of data and
       vague                                                                                               handle a few simple rules
October 06                        Case -Based Reasoning                                17   October 06                       Case -Based Reasoning                       18
                         Expert Systems                                                                    Expert Systems
             • others handle large collections of complex rules, but do not
               perform well when accessing data
                                                                                   • because expert systems are highly
                                                                                     specialised, static systems, they can be
     – many do not learn
             • makes them obsolete within short time of development                  extremely brittle when presented with novel
             • if needed in forseeable future, will require maintenance of           problems or situations.
               knowledge base and rest of system
                                                                                   • Expert system shells are environments for
• receives input describing problem in field of                                      creating expert systems
  expertise
                                                                                        – wide variety of expert system shells are
     – uses its inference to extract appropriate information                              commercially available, tend to be very
       from its KB to produce an answer                                                   expensive.

October 06                        Case -Based Reasoning                       19   October 06                       Case -Based Reasoning                  20




                   Expert System Shells                                                              Expert System Shells
     – CLIPS, a shell developed by NASA and written in                                          • knowledge base - provides all the rules
       ANSI C, is available for free at:                                                        • inference engine - controls overall execution of rules
             • http://www.ghgcorp.com/clips/download/                                   – program may consist of rules, facts and objects
• CLIPS                                                                            • CLIPS applications
     –   complete environment for developing ES                                         – ES for wheelchair selection
     –   C Language Integrated Production System                                        – for people with MS
     –   shell - portion of CLIPS which performs inference                              – involves examination of number of characteristics
     –   provides the basic elements of expert system                                           • ambulation status
             • fact list & instance list - global memory for data                               • length of diagnosis



October 06                        Case -Based Reasoning                       21   October 06                       Case -Based Reasoning                  22




                   Expert System Shells                                                              Expert System Shells
             • funding sources                                                                  • Insurance
     – few experts so system developed to aid in process                                        • Mobility and comfort
     – from the therapists standpoint                                                           • image
             •   environments                                                      • have Patient database, patient’s needs and
             •   transport of wheelchair
             •   distance to be traversed
                                                                                     constraints
             •   caregiver status                                                  • wheelchair database
             •   current wheelchairs (consider modification)                       • conduct search to provide solution set and
     – from the patients standpoint                                                  explanation
             • cost
                                                                                   • 3rd Conference on Clips (website)
October 06                        Case -Based Reasoning                       23   October 06                       Case -Based Reasoning                  24

								
To top