Case Adaptation

1. Case Adaptation Sources: Chapter 8 www.iiia.csic.es/People/enric/AICom.html www.ai-cbr.org

2. New problem Selected case Solution Adaptation knowledge Adaptation Adaptation

3. Classes of Adaptation • No-adaptation • Transformational Analogy • Substitution Adaptation • Feedback based • constraint based • Compositional adaptation • Generative Solution Adaptation • Transformational Analogy • Derivational Analogy

4. No Adaptation • For classification/diagnosis tasks • If C is the NN for a new problem P then • class(P)  class(C) • If {C1,…,Ck} are the k-NN for a new problem P then • Class(C)  F({C1,…,Ck}) • For example: F({C1,…,Ck}) = “majority class among • {C1,…,Ck}”

5. Substitution Adaptation • Let C = (P,S); A problem P and a solution S • Adaptation problem: • Given • A problem P’ • A case C such that P is similar to P’ • Search a substitution  such that (S) solves P’ •  corresponds to an application of a rule transforming parts of the case (so it is not a substitution in the traditional sense)

6. Example Support for PC sale: • Cases are configuration episodes of PCs • User specifies his/hers requirements • System selects best PC (e.g., using CCBR) and change some components Example rules (Substitutional Adaptation): If (query.application = ‘database’ and case.diskSpace < 2GB) then target.diskSpace4GB

7. Example (2) Example rules (Substitutional Adaptation): If (query.application = ‘games’ and case.application ‘games’) then AddObjecttarget.addJoystick AddObjecttarget.addSoundCard Other rules to configure joystick and sound

8. CaseB (old) CaseA (new) Car type: sport Color: red Seating: 2 Valves: 48 Type: 5.7L Model name: name1 Price: 200,000 Year: 2003 Feedback: not successful Cause: price is too high Car type: sport Color: red Seating: 2 Valves: 48 Type: 5.7L Model name: name1 Price: 200,000 Year: 2003 Feedback: successful Retrieve Copy Adapt CaseC (adapted) Car type: sport Color: red Seating: 2 Valves: 40 Type: 3.6L Model name: name 2 Price: 150,000 Year: 2000 Feedback: successful Substitutional Feedback-based Rule: if price is too high then model  previous model

9. Substitutional Constraint-based CaseB (old) CaseA (new) Case ID: 456 Speed: high Price: middle Usage: sport Antitheft performance: middle Model Name: Toyota Sedan 07 Price: 10,500 Antitheft system: Product A Case ID: 123 Speed: high Price: middle Usage: sport Antitheft performance: high Model Name: Toyota Sedan 07 Price: 10,500 Antitheft system: Product A Retrieve Copy adapt Rule: if need higher Antitheft performance and Antitheft System = product A then Antitheft System product B Price  Price + 500 CaseC (adapted) Case ID: 123 Speed: high Price: middle Usage: sport Antitheft performance: high Model Name: Toyota Sedan 07+ Price: 11,000 Antitheft system: Product B

10. 1 2 n … (P’,S’) (P,C) Compositional Adaptation • Let C = (P,S); A problem P and a solution S • Adaptation problem: • Given • A problem P’ • A case C such that P is similar to P’ • Search a sequence of substitutions 1, …, nsuch that: S’ is a solution for P’

11. 1 2 n … (P’,S’) (P,C) Adaptation Operators (2) • Uses rule-based systems during adaptation • Roles of operators/rules: • General knowledge about the domain Adaptation knowledge