Download presentation
Presentation is loading. Please wait.
Published byKristopher McCarthy Modified over 8 years ago
1
20. september 2006TDT55 - Case-based reasoning1 Retrieval, reuse, revision, and retention in case-based reasoning
2
20. september 2006TDT55 - Case-based reasoning2 Introduction CBR Influenced by cognitive science Usage of remindings (”This reminds me of something I’ve seen before”) An important issue is how closely CBR systems should mirror how humans think
3
20. september 2006TDT55 - Case-based reasoning3 Introduction The steps of the CBR cycle Retrieval in CBR → Fetches previous cases that are assumed to be able to contribute to solve the target problem Reuse → Suggests a solution for the target-case from the solutions of the retrieved cases, possibly with an adaption process to fit the target-case better Revision → Evaluates the chosen solution with respect to degree of success Retention → The product of the most recent problem-solving episode is incorporated into the system’s knowledge
4
20. september 2006TDT55 - Case-based reasoning4 Retrieval in CBR Similarity assessment Surface features: the features given as a part of the case description Similarity-based retrieval is retrieval based on similarity of the surface features Ineffective to scan all cases in the base → Foot-print based retrieval → Validation
5
20. september 2006TDT55 - Case-based reasoning5 Retrieval in CBR Retrieval performance The solution quality is as important as the retrieval speed Problems that may influence the quality: → inadequate similarity measures → noise → missing values in cases → unknown values in the description of the target problem → the heterogenity problem – different attributes are used to describe different cases Work on how to solve this problem: → making the similarity measure be the subject of an adaptive learning process → guiding by domain knowledge
6
20. september 2006TDT55 - Case-based reasoning6 Retrieval in CBR Alternatives to similarity-retrieval: Adaption-guided retrieval → Retrieval of the cases which are easiest to adapt Diversity-conscious retrieval → Combines similarity and diversity measures to distinguish between cases of great similarity. Compromise-driven retrieval → A case is more acceptable than another if it is closer to the user’s query and it involves a subset of the compromises that the other case involves.
7
20. september 2006TDT55 - Case-based reasoning7 Retrieval in CBR Alternatives to similarity-retrieval: Order-based retrieval → Combine preferred values with preference information such as max and min values, and values that the user would prefer not to consider. Explanation-oriented retrieval → The goal is to explain how the system reached its conclusions. The easiest way of doing this is to use the explanation of the most similar case.
8
20. september 2006TDT55 - Case-based reasoning8 Reuse and revision in CBR Reuse can be as simple as returning the most similar case, but significant differences in target problem vs. retrieved case → need for adaption Adaption methods: Substitution adaption → exchanges parts of the retrieved solution Transformation adaption → changes the structure of the retrieved solution Generative adaption → derives the new solution by repeating the method used to derive the solution of the retrieved case
9
20. september 2006TDT55 - Case-based reasoning9 Retention in CBR The simplest form of retention is to just save the problem case and its solution as a new case The utility problem: As the case-base grows, every new case will not lead to a lot of new information (overlaps other cases), but will increase the searching time just as much Solution in general: → Delete harmful cases from the case base Solution in CBR: → Use a competence-model to decide each case’s contribution to the total problem solving competence
10
20. september 2006TDT55 - Case-based reasoning10 Retention in CBR Case-base maintenance Insert two new steps into the CBR cycle: Review – checks the quality of the system knowledge Restore – chooses and executes maintenance operations Categorization of maintenance policies: → how they gather data relevant to maintenance decisions → how they determine when to trigger maintenance operations → the types of maintenance operations available → how the maintenance operations are executed
11
20. september 2006TDT55 - Case-based reasoning11 Conclusions There is a significant amount of ongoing research on this subject A lot of the research is motivated by awareness of the limitations of the traditional approach
Similar presentations
© 2024 SlidePlayer.com. Inc.
All rights reserved.