Presentation is loading. Please wait.

Presentation is loading. Please wait.

Susan Craw Room SAS B18a Case Based Reasoning Advanced Knowledge Based Systems.

Similar presentations


Presentation on theme: "Susan Craw Room SAS B18a Case Based Reasoning Advanced Knowledge Based Systems."— Presentation transcript:

1 Susan Craw Room SAS B18a s.craw@comp.rgu.ac.uk http://www.comp.rgu.ac.uk/staff/smc/teaching/kbp3/ Case Based Reasoning Advanced Knowledge Based Systems Module CM4023

2 © The Robert Gordon University, Aberdeen 2 How do we solve problems? By knowing the steps to apply from symptoms to a plausible diagnosis But not always applying causal knowledge diseases cause symptoms symptoms do not cause diseases! How does an expert solve problems? uses same book learning as a novice but quickly selects the right knowledge to apply Heuristic knowledge (rules of thumb) I dont know why this works but it does and so Ill use it again! difficult to elicit

3 © The Robert Gordon University, Aberdeen 3 Another way we solve problems? By remembering how we solved a similar problem in the past This is Case Based Reasoning (CBR)! memory-based problem-solving re-using past experiences Experts often find it easier to relate stories about past cases than to formulate rules

4 © The Robert Gordon University, Aberdeen 4 Problems we solve this way Medicine doctor remembers previous patients especially for rare combinations of symptoms Law English/US law depends on precedence case histories are consulted Management decisions are often based on past rulings Financial performance is predicted by past results

5 © The Robert Gordon University, Aberdeen 5 Retain Review Adapt Retrieve Database New Problem SimilarSolution CBR Solving Problems

6 © The Robert Gordon University, Aberdeen 6 CBR System Components Case-base database of previous cases (experience) episodic memory Retrieval of relevant cases index for cases in library matching most similar case(s) retrieving the solution(s) from these case(s) Adaptation of solution alter the retrieved solution(s) to reflect differences between new case and retrieved case(s)

7 © The Robert Gordon University, Aberdeen 7 R 4 Cycle REUSE propose solutions from retrieved casesREVISE adapt and repair proposed solution RETAIN integrate in case-baseRETRIEVE find similar problems

8 © The Robert Gordon University, Aberdeen 8 Applications Failure prediction ultrasonic NDT of rails for Dutch railways water in oil wells for Schlumberger Failure analysis Mercedes cars for DaimlerChrysler semiconductors at National Semiconductor Maintenance scheduling Boeing 737 engines TGV trains for SNCF Planning mission planning for US navy route planning for DaimlerChrysler cars

9 © The Robert Gordon University, Aberdeen 9 More Applications e-Commerce sales support for standard products sales support for customised products Personalisation TV listings from Changing Worlds music on demand from Kirch Media news stories via car radios for DaimlerBenz Re-Design gas taps for Copreci Formulation (recipes) rubber for racing tyres for Pirelli tablets for AstraZeneca

10 © The Robert Gordon University, Aberdeen 10 Whats in a Tablet? surfactant aids wetting and dissolution of drug disintegrant allows rapid break down after swallowing lubricant enables it to come out of the die binder makes it cohesive to hold together filler provides bulk to be large enough to handle and compress (~65%) drug active ingredient (~25%)

11 © The Robert Gordon University, Aberdeen 11 Tablet Formulation Problem Given: physical and chemical properties of a drug desired dose Knowing: properties of available excipients Goal: choose 5 excipients and their quantities which achieve the desired mechanical and chemical properties of the tablet Solution fillerDCP92.3% binderGEL 2.1% lubricantMGS 1.0% disintegrantCRO 2.1% surfactantSLS 0.3%

12 © The Robert Gordon University, Aberdeen 12 Tablet Formulation Knowledge Get-Insoluble-Filler IF: Reqd-Filler-Solubility has value Insoluble Filler is-on Filler-Agenda Solubility has value Sol in Filler Slightly-Soluble has value Slightly-Soluble Sol < Min-Val (Slightly-Soluble) THEN refine Filler to be Filler in Formulation Remove-Excessive-Fillers IF: Filler is-on Filler-Agenda Max-Level of Filler is Level Filler-Concentration has value Conc Conc > Level THEN... Heuristics Try to balance physical properties with stable excipients to achieve a tablet with viable properties Drug Properties Excipient Properties Drug/Excipient Stabilities Chemical Relationships Physical Relationships

13 © The Robert Gordon University, Aberdeen 13 Retain Review Adapt Retrieve Database Dose & Properties of New Drug SimilarSolution tablets of similar dose whose drugs have similar properties formulations for existing tablets soluble drug? => insoluble filler larger dose? => less filler CBR for Tablet Formulation

14 © The Robert Gordon University, Aberdeen 14 CBR Assumption New problem can be solved by retrieving similar problems adapting retrieved solutions Similar problems have similar solutions ? S SS S S S S S S P P P P P P P P P X

15 © The Robert Gordon University, Aberdeen 15 Why was filler X chosen? The tablet in the case-base whose drug properties are most similar dose is most similar is Drug-Y-50 and its filler is Z However adaptation is needed because of a significant difference the stability of Z with the new drug is much lower Adaptation proposes filler X instead: greater stability with new drug similar properties to Z

16 © The Robert Gordon University, Aberdeen 16 Why was filler X chosen? The tablet in the case-base whose drug properties are most similar dose is most similar is Drug-Y-50 and its filler is Z However adaptation is needed because of a significant difference the stability of Z with the new drug is too low Adaptation proposes filler X instead: greater stability with new drug similar properties to Z

17 © The Robert Gordon University, Aberdeen 17 CBR Knowledge Containers Cases lesson to be learned context in which lesson applies Description Language features and values of problem/solution Retrieval Knowledge features used to index cases relative importance of features used for similarity Adaptation Knowledge circumstances when adaptation is needed alteration to apply

18 © The Robert Gordon University, Aberdeen 18 Corporate Memory Cases from database, archive,... Issues case bias? currency? coverage? description language e.g. agreement on terms Case-base cannot contain all formulations good coverage prototypical and exceptional cases Opportunity for multiple sources several expert formulators shared knowledge across companies

19 © The Robert Gordon University, Aberdeen 19 Case Representation feature-value representation Problem drug properties and dose Solution excipients and their amounts Extra tablet properties constrained features of resulting tablet

20 © The Robert Gordon University, Aberdeen 20 CBR Tool C4.5 Index K Nearest Neighbour Similarity Matching progress of retrieval Database Relevant Cases Most Similar Cases Vote Tcl for adaptation Gshadg hjshfd fhdjf hjkdhfs hjdshfl hfdjsfhdjs hjdhfl hsdfhl hd hdjsh hjsdkh hfds hhfkfd shk Gshadg hjshfd fhdjf hjkdhfs hjdshfl hfdjsfhdjs hjdhfl hsdfhl hd hdjsh hjsdkh hfds hhfkfd shk

21 © The Robert Gordon University, Aberdeen 21 Nearest Neighbour Retrieval Retrieve most similar k-nearest neighbour k-NN like scoring in bowls or curling Example 1-NN 5-NN

22 © The Robert Gordon University, Aberdeen 22 How do we measure similarity? Distances between values of individual features problem and case have values p and c for feature f Numeric features f(problem,case) = |p - c|/(max difference) Symbolic features f(problem,case)= 0 if p = c = 1 otherwise Distance is (problem,case) weighted sum of f(problem,case) for all features Similarity(problem, case) = 1/(1+ (problem,case))

23 © The Robert Gordon University, Aberdeen 23 HighLow 200 0 100 300 Decision Trees as an Index Solubility? Dose?? ? ? ? lowhigh <200>200

24 © The Robert Gordon University, Aberdeen 24 Case Retrieval Typical implementation e.g. Case-Base indexed using a decision-tree Cases are stored in the index leaves… from these the most similar are retrieved using similarity matching

25 © The Robert Gordon University, Aberdeen 25 Why do we want an index? Efficiency if similarity matching is computationally expensive Pre-selection of relevant cases some features of new problem may make certain cases irrelevant... despite being very similar HighLow 200 0 100 300

26 © The Robert Gordon University, Aberdeen 26 Case Retrieval Parameters Selection of features inducing decision tree index Parameters to induce decision tree index Number of best-matches retrieved by similarity measure Weights for features similarity matching

27 © The Robert Gordon University, Aberdeen 27 Are CBR Systems Easy to Develop? Retain Review Adapt Retrieve Database Similar Past Cases Similarity Knowledge Adaptation Knowledge OK? Not Necessarily!

28 © The Robert Gordon University, Aberdeen 28 CBR System Database of previous formulations Similarity Matching Index Case-base Acquiring Knowledge Adaptation CBR A Profiles Adaptation rules

29 © The Robert Gordon University, Aberdeen 29 Learning Case-base inserting new cases into case-base updating contents of case-base to avoid mistakes Retrieval Knowledge indexing knowledge features used new indexing knowledge similarity knowledge weighting new similarity knowledge Adaptation knowledge

30 © The Robert Gordon University, Aberdeen 30 CBR Resources CBR Tools ReCall (www.isoft.fr), Orenge (www.tecinno.com) Kaidara (www.kaidarausa.com) CBR Websites www.ai-cbr.org www.aic.nrl.navy.mil/~aha/ www.scms.rgu.ac.uk/research/kbs/kacbd/ CBR Conferences ICCBR01: www.iccbr.org/iccbr01/ UK-CBR01: www.ai-cbr.org/ukcbr5/ ECCBR 2002: www.scms.rgu.ac.uk/eccbr2002/

31 © The Robert Gordon University, Aberdeen 31 Reading Useful texts (Kolodner 1993, Aamodt & Plaza 1994, Thompson 1997) Our papers Case-Based Design for Tablet Formulation. Craw, Wiratunga & Rowe. Proc. 4th European Workshop on CBR, p358-369, Springer, 1998. Self-Optimising CBR Retrieval. Jarmulak, Craw & Rowe. Proc 12th Int Conf on Tools with AI. IEEE Press, 2000. Using Case-Base Data to Learn Adaptation Knowledge for Design. Jarmulak, Craw & Rowe. Proc 17th Int Joint Conf on AI. AAAI Press, 2001. Also see http://www.scms.rgu.ac.uk/research/kbs/kacbd/

32 © The Robert Gordon University, Aberdeen 32 CBR vs Rule-based KBS Rule-based a rule is generalised experience applies to range of examples currently do not learn as they solve problems knowledge acquisition bottleneck Case-based reasoning cases include both prototypical cases and exceptions indexing, similarity and adaptation control effectiveness domain does not have an effective underlying theory learning updates case-base knowledge acquisition? retrieval and adaptation knowledge

33 © The Robert Gordon University, Aberdeen 33 Pros & Cons of CBR Advantages solutions are quickly proposed derivation from scratch is avoided domains do not need to be completely understood cases useful for open-ended/ill-defined concepts highlights important features Disadvantages old cases may be poor library may be biased most appropriate cases may not be retrieved retrieval/adaptation knowledge still needed

34 © The Robert Gordon University, Aberdeen 34 Summary CBR Cycle retrieve, reuse, revise, retain Knowledge containers case-base and description language retrieval and adaptation knowledge CBR tools to ease development of CBR systems C4.5 index and k-NN retrieval adaptation? Knowledge acquisition case knowledge can be easy retrieval/adaptation knowledge may not be easy


Download ppt "Susan Craw Room SAS B18a Case Based Reasoning Advanced Knowledge Based Systems."

Similar presentations


Ads by Google