Date: Thu, 28 Mar 2002 10:23:53 -0600 (CST) Message-Id: <200203281623.KAA09906@lucca.unl.edu> From: "Berthe Y. Choueiry" To: cse976-ml@cse.unl.edu Subject: Discussion on CBR. Content-Length: 1478 A few ideas inspired from Praveen's presentation and Cory's minutes. - I would like to stress the distinction in the paper between a process (cycle) and a task (there are four of them: retrieve, reuse, revise and retain). - Regarding CBR vs Expert Systems (ES), note that there are various approaches to deal with fuziness in ES. Certainty factors and fuzzy rules are just the most obvious examples. The major difference between CBR and ES in my opinion is that the former stores entire cases *and* their solutions while the latter does not. Note that both rely on background knowledge (in the form of rules or others). - Regarding neural nets. Remember that training a neural net requires a large data set, which you don't necessarily have in CBR. - Regarding CBR is more intelligent, more like human mind. It is definitely more like some aspects of human mind, but it does not entice us to "think out of the box," which in my opinion is also a feature of rational reasoning. - Regarding situations with few similarities. This is the major bottleneck of induction-based techniques. - Regarding the relation between CBR and psychology. Yes, CBR is one example of how advances in psychology research (more specifically cognitive science) directly benefits automated problem solving. It is very important to keep such channels open and intensify activities in these directions. Thanks, -Berthe Y. Choueiry choueiry@cse.unl.edu ========================= Date: Thu, 28 Mar 2002 12:09:52 -0600 From: "V. Praveen Guddeti" To: "Berthe Y. Choueiry" cc: cse976-ml@cse.unl.edu Subject: Re: CBR minutes In the comments of Tibor he raised the question of storing unsolved ones. There are no unsolved problems. The CBR may give wrong solutions. These failures of CBR are retained in the failure memory during Case Revision and made use of during the Derivational reuse phase of Case Reuse. On evaluating that the solution is wrong the CBR system invokes other methods to solve the problem and remembers this solution. That is why we need to have Integrated Approaches in a CBR system. For example integrating rule-based methods and CBR. Praveen. ============================ Date: Thu, 28 Mar 2002 13:36:25 -0600 (CST) Message-Id: <200203281936.NAA10130@lucca.unl.edu> From: "Berthe Y. Choueiry" To: vguddeti@cse.unl.edu CC: cse976-ml@cse.unl.edu Subject: Re: CBR minutes Praveen> In the comments of Tibor he raised the question of storing Praveen> unsolved ones. There are no unsolved problems. what if a problem *does* not have a solution? it would be good to remember this too, perhaps in some concise, reusable form. I don't see how the CBR `paradigm' prevents us from doing so. What do you think? -Berthe Y. Choueiry choueiry@cse.unl.edu ============================ Date: Thu, 28 Mar 2002 15:56:37 -0600 (CST) From: "V. Praveen Guddeti" To: "Berthe Y. Choueiry" cc: vguddeti@cse.unl.edu, cse976-ml@cse.unl.edu Subject: Re: CBR minutes By a solved problem I had assumed that either it has a solution if it is solvable or something stating that it has no solution if it is not solvable.So stating that a problem is not solvable is a type of "solution/answer" for that problem and so is taken care of by the CBR. By unsolved problem I had thought that you did nothing for that problem if no right solution was found for it. Praveen. ============================ Date: Thu, 28 Mar 2002 17:34:33 -0600 (CST) From: "Berthe Y. Choueiry" To: vguddeti@cse.unl.edu CC: cse976-ml@cse.unl.edu Subject: Re: CBR minutes > By a solved problem I had assumed that either it has a solution if > it is solvable or something stating that it has no solution if it is > not solvable. agree, if we know nothing about a problem, *perhaps* there is nothing we should store.. even so, the fact that "so far we know nothing about the problem" can sometimes be a valuable information. At least, we don't even search the Case libraries for a match :-) OK, I won't add to this thread myself... -Berthe Y. Choueiry choueiry@cse.unl.edu ============================