Applications of Case-based Reasoning in Macau
Vong Chi Man Li Yi Ping Wong Pak Kin Mak Peng Un Vai Meng I*
Abstract
Case-based reasonging (CBR) is an emerging methodology in ArtificialIntelligence (AI). A case is a situation, episode, etc. remembered (stored) by acomputer, and the attempt to reuse it is the subject of case-based reasoning[3,4,5]. CBR reasons by means of past cases stored in its case library, byretrieving similar cases in the past matched against the user input problemsituation, and then using pre-defined adaptation knowledge to reuse the retrievedcase to fit current situation. This paper presents the basic concepts of CBR, itsreasoning cycle, the major classes of adaptation methods used in CBR and abrief introduction of case-based adaptation (CBA). In addition, severalapplications of CBR in Macau are presented in the paper.
1.Introduction
1.1 What is Case-based Reasoning?
At its most simple, CBR is based on the observation that, when we solve aproblem, we often base our solution on one that worked successfully for a similarproblem in the past. An example of CBR would be someone driving to work inthe morning. That person doesn't explicitly plan his route, he just takes theroute he usually does. If that person meets a traffic jam, he may remember howhe avoided that jam in the past. If that person tries a new route to avoid thetraffic jam and he is successful, he will remember it and perhaps use it in thefuture. Another example would be a student doing a statistics assignment. Whenhe sees a new problem, he will try to see if there is any similarity between thecurrent problem and the ones he learned in the class. If there exists a past problemof which the solution traces fit onto the current one, the student would apply thesame past method to solve the current problem. These examples show the majortwo steps in the process of CBR - retrieval and adaptation, which will bediscussed further.
A more formal definition for CBR: CBR is a simple problem-solvingparadigm that involves matching a current problem against problems that weresuccessfully solved in the past. The process can be augmented by adaptingsolutions so that they more closely match the current problem.
1.2 The advantages of Case-based Reasoning
CBR suggests a model of reasoning that incorporates problem solving,understanding and learning and integrates all with memory processes [3]. Theadvantages of CBR are briefly discussed below:
·Reusing old cases is advantageous in dealing with situations that alwaysrecur. With reference to old cases, complexities of solving novelsituations can be reduced.
·Traditional reasoning processes cannot recall a relevant case unless itunderstands the new situation it is in. This suggests that understandingor interpreting a situation is a necessry part of the reasoning cycle andboth a prerequisite to problem solving and a co-requistie during problemsolving. But the need for problem understanding is not specific to CBR.
·CBR does not require exact description of problem to proceed thereasoning process while traditional reasoning (rule-based reasoning)cannot work with incomplete problem description.
·CBR emphasizes the use of concrete instances over abstract operatorsbecause they can provide more guidance and operational knowledge insolving a new problem than can abstract operators. Furthermore, they canshow application and use of knowledge that abstract operators do not supply.
·CBR emphasizes manipulation of cases over composition,decomposition, and recomposition processes. Though sometimescomposition tasks still occur in reasoning, reasoning using concrete caseswill come at first and then composition of operators next.
·It is usually necessary to adapt an old solution to fit a new situationbecause no old case is ever exactly the same as a new one. Adaptationcompensates for the differences between an old situation and a newone, while most parts of the old case can still be reused.
·Learning of CBR occurs as a natural consequence of reasoning byaccumulating new cases. Successful past cases can provide informationor procedure to solve current problem while unsuccessful past casescan warn that certain undesired results, happened in the past, should beavoided in current situation.
·Evaluation of cases can be based on feedback and follow-up proceduresto judge whether the cases are useful and give contribution ot learning.If cases are learnt (stored) without any follow-up analysis and evaluation,the reasoning using past cases will be unreliable.
1.3 Methodology in Case-based Reasoning
As it was mentioned above, CBR consists of two main parts - retrieval andadaptation. But in fact and in detail, CBR is constituted with four RE's [5]:
1. retrieve - retrieve similar past case matched against current problem.
2. reuse - reuse to solve current problem based on solution of past case.
3. revise - revise the past solution if any contradiction occurs when applied to current problem.
4. retain - retain the final solution along with the problem as a case ifthe case is useful in the future.
Fig. 1 illustrates the whole process of CBR. When user inputs a problem,the problem is interpreted and converted as a new case into the specific formatof the reasoning system. Then the converted new case enters the stage ofRETRIEVAL where the new case is matched against the previous cases in thecase library of the reasoning system. The retrieved case and the new case areboth passed to next stage REUSE where the solution part of the retrieved caseis applied to the new case, with the guidance of adaptation knowledge. Theapplication of old solution involves substitution of solution features, structuralmodification of the solution and/or derivational replay of solution. These methodswill be discussed at next section. After this stage, now the new case is alongwith the adapted solution based on the old one. This adapted solution isconsidered as a suggested solution, which is still incomplete because it is adaptedaccording to the requirement of new case and this solution may haveinconsistency among its solution parts. To ensure the adapted solution isconsistent, it will be passed to next stage REVISE, where the adapted solutionwill be further adapted based on the user feedback and additional meta-adaptationknowledge called repair. This final adapted solution is then returned as aconfirmed solution to the new case but this does not come to an end. It wasdiscussed that CBR learns by accumulating new cases. However, should anynewly derived case be accumulated in the case library? The answer is no. CBR should only store the cases that can contribute to future reasoning of solutions,which could not be done only by the cases in current case library. In other words,if the cases in a case library are capable enough to cover the newly adapted solution,this new solution should not be stored in order to avoid redundancy andinconsistency. Otherwise, it should be stored for future use. This leads to the finalstage of CBR-RETAIN. The case considered as being able to contribute in thefuture is named "Learned case" and stored in the case library.
2.Adaptation Methods in Case-based Reasoning
In the past, most of the research and development concentrate on the stageof retrieval and similarity of problem pairs, so many different and efficientmethods for evaluating similarity of cases were developed. However very feweffort has been done on the complicated and more important part of CBR -adaptation [4,7,10,11,12,17,18]. Adaptation of CBR can be done by threedifferent methods, according to the complexity of the problem domain. Themethods are substitutional adaptation, transformational adaptation andderivational replay.
2.1 Substitutional Adaptation
Substitution is the process of choosing and installing a replacement forsome part of an old solution [3]. Substitution is usually done by reinstantiationand parameter adjustment.
Reinstantiation is the process to find another object of the same class toreplace the old object [3]. It is used when the frameworks of an old and newproblems are obviously the same but roles in the new case are filled differentlythan roles in the old one. In traveling planning, we want to travel from Macauto Shanghai. We define that "traveling way" is a class of the way to traveling,which simply consists of "Airplane", "Ship", "Train" and "Vehicle". A previouscase may record that the traveling was by train. However, new case requires alimited time constraint so that "Train" will be replaced / reinstantiated by"Airplane" to fulfill this constraint.
Figure 1.The case-based reasoning cycle
Parameter adjustment is a technique for interpreting values in a new solutionbased on those from an old one [3]. In parameter adjustment, changes inparameters in an old solution are made in response to differences betweenproblem specifications in an old and a new case. It is a two-step process. First,the old and new problem descriptions are compared and their differences areextracted. Then adaptation knowledge, ususally in the form of rules, are selectedbased on requirement and applied to the old solution to create a new one thatsatisfies the requirement of the new problem. Using the above traveling example,if time constraint is specified in a new problem, then traveling way will besubstituted by "Airplane". Furthermore, the parameter of the new solution "costof traveling" will be adjusted because the ticket fee for airplane is more expensivethan the fee for ship. This is adjusted by multiplying a certain ratio to the initialticket fee to get a consistent new solution.
2.2 Transformational Adaptation
Substitution is appropriate if there already exists an item or concept thatcan be substituted for some inappropriate value. But it cannot be used if theitem required does not yet exist. Nor can it be used for insertions or deletions ofextra items in an old solution. [3]
To compensate the inability of adding to or removing items from an existingsolution to become a new solution, another kind of adaptation method,transformational adaptation, is proposed. Under this objective, this methodmainly deals with the reconstruction of solution structure, augmented withsubstitutioani techniques to adapt a solution.
Using the traveling example, if a traveling by train was adapted to fit thelimited time constraint specified in the current problem, then the "way oftraveling" will be filled in by "Airplane". At this time, another field "AirportTax" should be inserted as a part of solution in the record to specify the amountof tax to pay, which does not need to specify if traveling way is by ship, train orvehicle. Inversely, if an airplane traveling was adapted to a ship, train or vehicletraveling, the field "Airport Tax" in the old solution should be removed tomaintain the consistency of the new solution.
2.3 Derivational replay
Derivational replay [13,19,20] is the process of using the method ofderiving an old solution or solution piece to derive a solution in the newsituation. Rather than storing and directly reusing a solution itself, CBRcan store a trace of how that solution was generated and replay it in thenew situation. A trace is a sequence of reasoning steps with justification,similar to the operators in traditional rule-based reasoning. So it can alsobe considered as a chaining of operators. When the solution is replayed tosolve future problems, the replay process can directly take into accountdifferences between the old and new situations. Since every step in thesolution trace is along with explanation on why this step is taken, memoryorganization of case structure becomes very complex. So, this method isnot recommended for simple and easy task domain. In complex task domainssuch as route planning, design and synthesis tasks, this adaptation methodseems to be the most appropriate so far.
Example: In a previous traveling case by train from Macau to Shanghai, it records the following:
──────────────────────────────────────────
From To By Reason Justification
──────────────────────────────────────────
Macau Zhuhai Vehicle (vehicle to Guangzhou) Always necessary
Zhuhai Guangzhou Vehicle (train to Shanghai) if by train or vehicle
Guangzhou Shanghai Train (Goal) if by train
──────────────────────────────────────────
Old case organized for derivational replay
|
Then the current problem is to find a traveling path by train from Macau toBeijing. The CBR process will consider the first solution step. Because the firststep is always necessary according to the justification, no matter which travelingway is to take, the new solution will consist of first step from the old solution.Then, the second step is taken into consideration whether it should be included,now the traveling is by train. So, we need to go to Guangzhou to take a train andhence the second step is also included in the new solution. For the third step, itis also included because it is the action to travel to Beijing but simple substitutionis needed to modify the destination from Shanghai to Beijing and the new solutionwill become:
──────────────────────────────────────────
From To By Reason Justification
──────────────────────────────────────────
Macau Zhuhai Vehicle (vehicle to Guangzhou) Always necessary
Zhuhai Guangzhou Vehicle (train to Shanghai) if by train or vehicle
Guangzhou Beijing Train (Goal) if by train
──────────────────────────────────────────
New case generated by applying derivational replay
|
This example is simple and it seems to be more or less the same as done insubstitutional adaptation. However, it involves synthesis of solution pieces byconsidering the solution pieces one by one to dertermine whether they shouldconstitute the new solution (transformational adaptation), plus necessary valuesubstitution (substitutional adaptation).
3.Case-based Adaptation
Adaptation is an important process in CBR as discussed above.Usually adaptation knowledge is in the form of rules specifying,under a certain situation,how the value of a feature is modified,or some feature should be inserted or removed to generate a solution for the new problem.At this time,know-how adaptationknowledge can be applied because the problem domain is simple and wellunderstood so that features to adapt can be easily allocated by comparing differencesbetween the new problem and its past similar case. However, in many real worldproblem domains, especially planning and design, there are no universal rules foradapting a feature, or the rules are too abstract to use. Then know-how adaptationknowledge can only be applied based on expert experience. The selection of whichrule from a set of adaptation rules for a feature may change under different situations.Then another source of adaptation knowledge is necessary to guide the selectionof suitable rule from a set adaptation rules applied to a feature. Because the selectionis based on expert experience, another level of CBR could be established to supportthe selection of adaptation rules. This method of integrating two levels of CBR iscalled case-based adaptation (CBA) [11,16,17,18].
In case-based adaptation, guidance for adaptation, both what to adapt andhow, is given by a previous case [3]. The case supplying this adaptationinformation is called adaptation case. When potential problems with solutionsare identified, it is not always clear exactly what should be fixed to alleviate theproblem. If blindly applying adaptation rules in a planning or design domain,the newly generated solution may not be idiosyncratic although it satisfies theentire specification requirement from user. However, previous cases to guideadaptation can promote novel adaptations leading to creative solution.
Domain expert's domain knowledge of how adaptation is performed, includingwhich features to adapt and which adaptation rules are used, etc., is captured intothe adaptation cases, presumably provides adaptation cases. After initial installationof adaptation case library, as time passes by, the adaptation case library will befilled with more and more adaptation cases by repeatedly using of the CBR systemby users. This provides another way to acquire expert knowledge that is not easy tocapture via communication into production rules. Of course, the main objective ofcase-based adaptation is to provide a more convenient and consistent method toperform case adaptation in some complex planning or design domains.
4.Methodologies for Building CBR systems
Full acceptance of CBR by industry depends on establishing softwaredevelopment methodologies for CBR, to define how to organize and developCBR projects. Lessons from CBR applications form a foundation for definingsuch methodologies [9]. One fundamental principle revealed by manyexperiences is the value of an iterative development process. Since CBR systemscan provide useful results even with a partial case library, systems can be set up with a set of seed cases that is augmented as gaps revealed during use. Basically the process for developing a CBR application follows three phases:
Case-based design: a general representation for cases is developed using sources at hand (e.g., documentation, database records, and written accounts by experts). This is accomplished by a coordinated effort involving users, managers, and the developers of the system. A lexicon of terms used to describe problem features, the selection of appropriate features for case indexing and the specification of database schemas used to store cases are defined at this stage.
Inital case-based development: a small case base is initially developed to provide a base for the application. The initial case base is then extensively reviewed both by developers and users and is repeatedly refined until a valid case structure and case base covering a large portion of the application area is complete.
Ongoing development and maintenance: the case base is managed in a traditional manner by a database administrator. Statistical quality control techniques may be used to monitor case accuracy and utility. Current research of maintenance of case library focuses on the RETAIN stage of CBR process which skip the storing of a case if the case is not covering any novel situation.
5.Application Examples of CBR in Macau
CBR can be applied in a variety of different task contexts, including retrieval-only, classification, diagnosis, planning, design and decision-support.The following are some of applications of CBR in Macau.
5.1 Case-based Reasoning in Airport Terminal Area Planning
During the daily operations of an airport terminal, many decisions have to bbe made on how to allocate facilities during a certain time in order to avoid passenger congestion. This job falls in the category of airport planning [1,2].Airport planners and managers are frequently confronted with two problems:
·How to determine the number of various major functional elements of an airport terminal building that are needed to handle a given passenger throughput, and, conversely.
·How to determine the passenger throughput capability of existing airport terminal facilities.
Figure 2 depicts the passenger flows for arrival and departure and the major functional facilities1 considered in this project.
To solve these frequent problems, CBR is used to aid in the decision making process of an airport planner. The system helps the planner in deciding the number of facilities to allocate at certain times of the day given the number of peak hour passengers, number of available facilities, etc. The idea for applying CBR to this process came from a project developed in 1995 in the University of Macau. The project, ATA-Sim, is a simulator of airport operations that helps predicting future requirements of airport terminals in terms of facilities. This system is very robust but it is a simulation, it is more appropriate to be used in medium to long term planning. When the day-to-day operations of an airport are at stake, a system with a faster response time and a larger experience in the field of airport planning is needed; this system fills that gap very well.
5.2 Prediction of radio signal propagation for GSM networks
Another application area of CBR is in the prediction of radio signal propagation in mobile phone networks [1,2]. A mobile phone network is constituted of transmission elements distributed around a certain area. Each element has a certain transmission range determined by the location, height, signal strength, and type of transmission station and constrained by the obstacles that are in the path of the radio waves, atmospheric conditions, etc. Predicting future requirements for this kind of network is very important since the growth of a city is inevitable, new obstacles will prevent radio waves from propagating to certain areas, old obstacles may be removed, making stations in other areas unnecessary.
Figure 2.Passenger flow in an airport terminal
The addition of a new transmission station in a certain area may arise after verification that the signal strength in that area is no longer above a certain threshold value for acceptable communication. If, in the past, a similar situation occurred where the area had similar characteristics to the area in the current problem, past experience in solving such a problem can be applied in the solution of the current problem. CBR systems use prior episodes to solve new problems, thus, this class of system can be applied to this problem.
Figure 3.A portion of a mobile phone network
The primary focus of the CBR system in this application will be to advise on the height of the transmission station, the downtilt2 of the antenna and the transmitting power of the station. These values will be features of cases in the system's case library. This case library will contain a series of previous experiences, each in the form of a case, including all of the stations that are in operation. A problem case is posed to the system by entering the beamwidth, the distance covered by the transmitting station and the average height of buildings in the area. The system will search its case library and return a case with the most similar situation.
The structure of a case will be divided into three different parts, the first will contain fields that will be entered and given to the system to start its search.This is the problem description part. The recommendation, or solution, part will contain the fields that provide necessary information to establish a stationin the new area.Finally the third part will contain estimates and measures of signal strength in the area of influence of the new station.The proposed case structure for the CBR system is listed in the table 1.
┌──────────────┬────────┬─────────────────────┐
│Field │Field's class3 │Explanation │
├──────────────┼────────┼─────────────────────┤
│Average height of buildings │Problem │Used to narrow the search procedure │
│in the area │ │of CBR and limit it to cases whose │
│ │ │areas have similar characteristics to the│
│ │ │problem area. It would not be of much │
│ │ │interest if the system recommended the │
│ │ │height of the antenna to be 100 meters │
│ │ │when the maximum height of a │
│ │ │building in the area was 50 meters. │
├──────────────┼────────┼─────────────────────┤
│Beamwidth │Problem │The transmitting beamwidth of the │
│ │ │antenna. │
├──────────────┼────────┼─────────────────────┤
│Distance covered │Problem │The maximum distance a mobile phone │
│ │ │unit can be from the station. │
├──────────────┼────────┼─────────────────────┤
│Station height │Solution │The height the station is from the │
│ │ │ground. │
├──────────────┼────────┼─────────────────────┤
│Antenna downtilt │Solution │The downward angle of the antenna. │
├──────────────┼────────┼─────────────────────┤
│Transmitting power │Solution │The power the station transmits the │
│ │ │signal. │
├──────────────┼────────┼─────────────────────┤
│Estimated signal strength │Estimation │The estimated strength of the signal │
│ │ │within the station's area of influence. │
├──────────────┼────────┼─────────────────────┤
│Measured signal strength │Measurement │The measured data indicating the │
│ │ │signal strength in the area. │
└──────────────┴────────┴─────────────────────┘
Table 1.Proposed case structure for GSM networds case-based reasoner
|
Note: Even though in the table above each field is classified under problem,solution, estimation, and measurement, all the fields are necessary to characterize a case. The class of each field will only be used as reference when the case- based reasoner is being consulted.
5.3 Case-based Adaptation applied to Hydraulic Circuit Design4
Hydraulic circuit design (HCD) is an industrial application domain from Electromechanical Engineering that supports drawing design of circuit used in oil hydraulic machines. A circuit design process involves sketching a possiblecircuit layout, selecting potential components and calculating component sizes or circuit limits. On completion of a circuit design, building prototypes and testing and evaluating the candidate system performance are carried out [14,15]. Selecting appropriate components and sketching the design layout, both are based on the experience of designer and most of the time, the selected layout and components are very similar to ones used in past circuit designs. In order to save effort of doing redundant and tedious task, HCD system is proposed to provide knowledge-based support to guide designer for the completion of circuit design.
Traditional HCD systems used rule-based reasoning technique to check the consistence of components in the circuit design. However, solving problem by analogy even fits into the current design domain. With the help of emerging CBR methodology, most of the problems mentioned above could be augmented at a certain level. For example, a similar layout could be retrieved and ready for user modification. Circuit design with little difference in parameter specification could be adapted automatically by the system, in just substituting the parameter values or even replacing certain components to fit the parameter specification. With CBR and CBA, the circuit design system could learn by accumulating new cases and repetitive training from expert users to refine its adaptability to new problem cases. Figure 4 briefly illustrates the structure of the program and interaction with users.
6.Conclusion
CBR is an emerging methodology in Artificial Intelligence. Its main purpose is to retrieve and reuse analogical past experience to solve current problem. In many domains, the rules of the domain have not yet been formalized or the formalization is so abstract that it is hard to use because there is so much variation in the way abstract concepts show themselves. CBR allows problem solving in such domains by basing reasoning on experience in the domain rather than on formal models, because cases capture associations between situations, solutions, and outcomes, capturing an informal understanding of the domain at a concrete level. For some tasks, CBR complements model-based or rule-based reasoning because cases can easily suggest previous solutions that are hard to retrieve under ill-defined concepts. When similar problem statements call for similar solutions, CBR has shown itself efficient and accurate when asked to solve problems similar to those it has been before. Since CBR provides a commonsense, intuitive model of problem solving, knowledge acquisition becomes easier than traditional reasoning paradigms. In planning and design domains, problems always recur. Much effort was wasted in repeating redundant jobs to find out a new solution while planning and design [4,8]. To save the effort from repeating redundancy, CBR could be employed to reuse past successful cases to derive a new solution for new problem. Furthermore, CBR is useful in aiding decision- making [6] when large amount of constraints have to be considered as discussed in the application example of Airport Planning. Then previous successful cases could be retrieved to present to users as reference, while previous failed cases as warning, which works exactly as the natural and intuitive human reasoning process.
From these advantages over traditional reasoning paradigms, CBR systems are easier to build and also efficient in solving similar problems, especially in complex domains such as planning and design.
Figure 4.Program structure for Hydraulic Circuit Design applied with case-based adaptation
Notes:
1 These are only a subset of all terminal facilities since this system is to be used to support the airport planner in his/her decision making process of how to allocate facilities on a day-to-day basis.
2 Downtilt is the downward inclination angle of the antenna.
3 The field's class is for reference regarding whether the field is a problem, solution or just a field containing estimated of measured signal strength.
4 This is an undergoing project in Faculty of Science and Technology, University of Macau.
References:
1. P. Galvão and C.M. Vong, "A step closer to modelling human reasoning Case-Based Reasoning", Macau Engineering Bulletin, No.3 December 1996,The Macau Institute of Engineers, pp. 54-56.
2. C.M. Vong, Case-Based Reasoning, Project report submitted as partialfulfillment of BSc. in Software Engineering.
3. J. Kolodner, Case-based Reasoning, Morgan Kaufman Publ., San mateo, CA, U.S.A., 1993.
4. F. Gebhardt, A. Voβ, W. Grthaer and B. Schmit-Belz, Reasoning withComplex Cases, Kluwer Academic Publishers, 1997.
5. I. Watson and F. Marir, Case-based Reasoning: A Review, The knowledge Engineering Review Vol.9, No.4, 1994.
6. M. Lenz, E. Auriol and M. Manago, "Diagnosis and Decision Support", in Case-Based Reasoning Technology: From Foundations to Applications, Springer, 1998.
7. W. Wilke, B. Smyth and P. Cunningham, "Using Configuration Techniques for Adaptation", in Case-Based Reasoning Technology : From Foundations to Applications, Springer, 1998.
8. K. B~:orner, "CBR for Design", in Case-Based Reasoning Technology : From Foundations to Applications, Springer, 1998.
9. R. Bergmann, and K. Althoff, "Methodology for Building CBR Applications", in Case-Based Reasoning Technology : From Foundations to Applications, Springer, 1998.
10. K. Hanney and M. Kean, "The Adaptation Knowledge Bottleneck: How to Ease it by Learning from Cases", in Case-Based Reasoning: Research andDevelopment (ICCBR 97), Springer, 1997.
11. D. Leake, A. Kinely and D. Wilson, "A Case Study of Case-Based CBR", in Case-Based Reasoning: Research and Development (ICCBR 97), Springer,1997.
12. L. Purvis and S. Athalye, "Towards Improving Case Adaptability with a Genetic Algorithm", in Case-Based Reasoning: Research and Development(ICCBR 97), Springer, 1997.
13. S. Bhatta and A. Goel, "An Analogical Theory of Creativity in Design", in Case-Based Reasoning: Research and Development (ICCBR 97), Springer, 1997.
14. K.K. Chan, Implementation of an Object-oriented ICAD system for Hydraulic Circuit, M.Sc Dissertation for the Department of Engineering, The University of Warwick, 1996.
15. P.K. Wong, T.P. Leung, C.W. Cheun and W.H. Chan, "Object-oriented CAD for electro-pneumatic squential circuit design in low cost automation", Proceeding of SEIKEN/IEEE symposium on emerging technologies & factory automation, Japan, Nov 1994, pp. 278-284.
16. B. Smyth and M. Keane, "Experiments on adaptation-guided retrieval in case-based design", in Case-Based Reasoning: Research and Development (ICCBR 95), Springer, 1995.
17. D. Leake, A. Kinley and D. Wilson, "Case-Based Similarity Assessment:Estimating Adaptability from Experience", Proceedings of the Fourteenth National Conference on Artificial Intelligence, AAAI Press, Menio Park,CA, 1997.
18. D. Leake, "Combining Rules and Cases to Learn Case Adaptation",Proceedings of the Seventeenth Annual Confemece of the Congnitive Science Society, 1995.
19. M. Veloso, "Flexible Strategy Learning: Analogical Replay of problem Solving Episodes", In Proceedings of AAAI-94, pp. 595-600.
20. P. Cunningham, D. Finn and S. Slattery, "Knowledge Engineering Requirements in Derivational Analogy", in Topics in Case-Based Reasoning(EWCBR 93), Springer, 1993.
* All are academic staff of Faculty of Science and Technology,University of Macau
