Subject: IJCAI-97 WORKSHOP REVIEW FORM Comments for Author Title: Ripple-Down Rationality: A Framework for Maintaining PSMs Number: 1. How RELEVANT is this paper to AI researchers? (mark one box) [x] Very relevant [ ] Moderately relevant [ ] Not relevant Please explain your rating: The paper claims that probelm-solving methods are not necessary and each problem-solving process can adequately expressed as abduction. 2. How SIGNIFICANT is this paper? [ ] Very significant [x] Moderately significant [ ] Not significant 3. How ORIGINAL is this paper? [ ] Very original [x] Moderately original [ ] Not original Please explain your rating: It was not so clear what is the orginality compared to earlier papers of the author? 4. Is this paper technically SOUND? [x] Yes [ ] Seems valid, but did not check completely [ ] Has minor errors [ ] Has major errors If there are errors, please identify them: 5. How well is this paper PRESENTED? Good Average Poor Overall organization [x] [ ] [ ] English [x] [ ] [ ] Readability [x] [ ] [ ] If there are presentational problems, please identify them: If I understood right you claim the following: PSMs that describe generic reasoning patterns does not exist in your framework (only as interpretation of an observer) but not as an internal structure you shape on top of low-level reasoning processes of your system. I have some questions: 1. If it is in the head of the observer why not using it as a high-level structuring of the reasoning process of the system to improve under- standability and maintainability? 2. As you write you do not have this meta-level reasoning (sek 3.2) of PSM that can be reused and help to built KBS more efficiently. Your reuse is on a very generic level (RDR and abduction). 3. Very interessting your comment on complexity problem of abduction. Finding efficient problem-solving via heuristic reasoning is precisely my concern for working with PSMs. They realize (under some assumptions) very efficient reasonig for limited subclasses of problems and with support of strong domain knowledge. Actually you self write the main counter argument that I would present against your approach: "Computationally tractable abductive inference algorithm typically make restrictive assumptions about the nature of the theory ..." I use my PSMs to define more specialized problem classes and I can therefore tune (adapt) their assumptions much more to the given specialized circumstances as assumptions that have to guarantee low complexity for general abduction which rules out most interesting cases. Do you see my point? 4. I only have a rough idea what you mean with: - a PSM is a set of BEST operators and - a BEST operator is a classification system Could you elaborate on this. Also for general understandability. Could you try to work out VT (propose & revise) in your framework. How would it look like? Where is the RDR systems that designs elevators as described in Yost document. Or is this the wrong question for you? ===================================== Comments for Program Committee Members Only (This section of the review will be withheld from the author eehm: Only if he is NOT a member of the organisation committee) Title: Ripple-Down Rationality: A Framework for Maintaining PSMs Number: 1. My recommendation is: [x] Accept [ ] Leaning to accept [ ] Leaning to reject [ ] Reject 2. How confident are you in your appropriateness as a referee for this paper? [ ] Very confident - I am an expert in this area. [x] Confident - I have a reasonable knowledge of this area. [ ] Fairly confident - I have some knowledge of this area. [ ] Not confident - I have no significant knowledge of this area. 3. If this paper is marginal or unusual in some way, can you comment on anything else that might help the Program Committee reach a decision? Reviewer: (please fill in your name here) Dieter Fensel --========================_47306234==_ Content-Type: text/plain; name="psm.6.author.txt"; charset="us-ascii" Content-Disposition: attachment; filename="psm.6.author.txt" Subject: IJCAI-97 WORKSHOP REVIEW FORM Comments for Author Title: Ripple-Down Rationality: A Framework for Maintaining PSMs Number: 1. How RELEVANT is this paper to AI researchers? (mark one box) [x] Very relevant [ ] Moderately relevant [ ] Not relevant Please explain your rating: The paper claims that probelm-solving methods are not necessary and each problem-solving process can adequately expressed as abduction. 2. How SIGNIFICANT is this paper? [ ] Very significant [x] Moderately significant [ ] Not significant 3. How ORIGINAL is this paper? [ ] Very original [x] Moderately original [ ] Not original Please explain your rating: It was not so clear what is the orginality compared to earlier papers of the author? 4. Is this paper technically SOUND? [x] Yes [ ] Seems valid, but did not check completely [ ] Has minor errors [ ] Has major errors If there are errors, please identify them: 5. How well is this paper PRESENTED? Good Average Poor Overall organization [x] [ ] [ ] English [x] [ ] [ ] Readability [x] [ ] [ ] If there are presentational problems, please identify them: If I understood right you claim the following: PSMs that describe generic reasoning patterns does not exist in your framework (only as interpretation of an observer) but not as an internal structure you shape on top of low-level reasoning processes of your system. I have some questions: 1. If it is in the head of the observer why not using it as a high-level structuring of the reasoning process of the system to improve under- standability and maintainability? 2. As you write you do not have this meta-level reasoning (sek 3.2) of PSM that can be reused and help to built KBS more efficiently. Your reuse is on a very generic level (RDR and abduction). 3. Very interessting your comment on complexity problem of abduction. Finding efficient problem-solving via heuristic reasoning is precisely my concern for working with PSMs. They realize (under some assumptions) very efficient reasonig for limited subclasses of problems and with support of strong domain knowledge. Actually you self write the main counter argument that I would present against your approach: "Computationally tractable abductive inference algorithm typically make restrictive assumptions about the nature of the theory ..." I use my PSMs to define more specialized problem classes and I can therefore tune (adapt) their assumptions much more to the given specialized circumstances as assumptions that have to guarantee low complexity for general abduction which rules out most interesting cases. Do you see my point? 4. I only have a rough idea what you mean with: - a PSM is a set of BEST operators and - a BEST operator is a classification system Could you elaborate on this. Also for general understandability. Could you try to work out VT (propose & revise) in your framework. How would it look like? Where is the RDR systems that designs elevators as described in Yost document. Or is this the wrong question for you? ===================================== ________________________________________________________ Title: Ripple-Down Rationality: A Framework for Maintaining PSMs Number: 6 1. How RELEVANT is this paper to AI researchers? (mark one box) [x] Very relevant [ ] Moderately relevant [ ] Not relevant Please explain your rating: The authors of this paper use as solution to all problem types (reasoning forms) abduction with an BEST operators. They distinguish several types of BEST operators, however the idea is that solution of a problem can be find using abduction and an application of BEST operators. Applying an operator can be considered as a classification problem. This idea leads to using a maintaining method for classification for maintaining PSMs (because a PSM is only abduction and the right BEST operation.) They use the technique ripple-down-rules extended with functions. (I am not happy with this view of problem solving methods, but this is just only my opinion. The methods for specific tasks becomes so superfluous.) 2. How SIGNIFICANT is this paper? [ ] Very significant [x] Moderately significant [ ] Not significant Please explain your rating: 3. How ORIGINAL is this paper? [x] Very original [ ] Moderately original [ ] Not original Please explain your rating: I think it is a further development in the lines of the ECAI96 paper. New is the consideration of using a method for classification for maintenance of PSMs. The particular view of PSM (abduction+BEST operator) enables them to use a classification method. 4. Is this paper technically SOUND? [ ] Yes [x] Seems valid, but did not check completely [ ] Has minor errors [ ] Has major errors If there are errors, please identify them: 5. How well is this paper PRESENTED? Good Average Poor Overall organization [ ] [x] [ ] English [x] [ ] [ ] Readability [ ] [x] [ ] If there are presentational problems, please identify them: The first part of the paper is well written and very readable. The example section is for me at several places difficult to understand completely. 6. Further comments, advice or explanations (Please be specific and constructive, especially with respect to any negative judgements above. Point to the section(s) where an error occurs, cite omitted references, etc.) Use as much space as you need. Terminology: I would never called hypotheses as IN. In you terms given OUT, the IN will be computed. So for me IN and OUT are a little bit confusing. In your first example, the assumption set is {xUp,yUp, cDown, gUp, gDown} Could you explain why you consider also assumptions (e.g. yUp) which can be derived from other assumption (xUp) as assumptions? A nice categorization of BEST operators. You have to introduce V a little bit more. This because of the example of Reiter with the negation of abnormals. A question about section 2.4 Complexity: Could you give requirements on the theory, or IN, or OUT such that abduction becomes doable? In the example section: I do not understand why the knowledge base of fig. 6 is translated (fig.8) to using partialAND's. Is the idea that this _a_ way of a translation of _the_ way of translation. It is difficult to see, what the technique for maintaining single classification is and what the extensions are that are needed for the application to maintaining PSM. Possibly there are no extensions at all. Subject: IJCAI-97 WORKSHOP REVIEW FORM Comments for Author Title: Ripple-Down Rationality: A Framework for Maintaining PSMs Number: 1. How RELEVANT is this paper to AI researchers? (mark one box) [x] Very relevant [ ] Moderately relevant [ ] Not relevant Please explain your rating: The paper claims that probelm-solving methods are not necessary and each problem-solving process can adequately expressed as abduction. 2. How SIGNIFICANT is this paper? [ ] Very significant [x] Moderately significant [ ] Not significant 3. How ORIGINAL is this paper? [ ] Very original [x] Moderately original [ ] Not original Please explain your rating: It was not so clear what is the orginality compared to earlier papers of the author? 4. Is this paper technically SOUND? [x] Yes [ ] Seems valid, but did not check completely [ ] Has minor errors [ ] Has major errors If there are errors, please identify them: 5. How well is this paper PRESENTED? Good Average Poor Overall organization [x] [ ] [ ] English [x] [ ] [ ] Readability [x] [ ] [ ] If there are presentational problems, please identify them: If I understood right you claim the following: PSMs that describe generic reasoning patterns does not exist in your framework (only as interpretation of an observer) but not as an internal structure you shape on top of low-level reasoning processes of your system. I have some questions: 1. If it is in the head of the observer why not using it as a high-level structuring of the reasoning process of the system to improve under- standability and maintainability? 2. As you write you do not have this meta-level reasoning (sek 3.2) of PSM that can be reused and help to built KBS more efficiently. Your reuse is on a very generic level (RDR and abduction). 3. Very interessting your comment on complexity problem of abduction. Finding efficient problem-solving via heuristic reasoning is precisely my concern for working with PSMs. They realize (under some assumptions) very efficient reasonig for limited subclasses of problems and with support of strong domain knowledge. Actually you self write the main counter argument that I would present against your approach: "Computationally tractable abductive inference algorithm typically make restrictive assumptions about the nature of the theory ..." I use my PSMs to define more specialized problem classes and I can therefore tune (adapt) their assumptions much more to the given specialized circumstances as assumptions that have to guarantee low complexity for general abduction which rules out most interesting cases. Do you see my point? 4. I only have a rough idea what you mean with: - a PSM is a set of BEST operators and - a BEST operator is a classification system Could you elaborate on this. Also for general understandability. Could you try to work out VT (propose & revise) in your framework. How would it look like? Where is the RDR systems that designs elevators as described in Yost document. Or is this the wrong question for you? ===================================== ________________________________________________________ Title: Ripple-Down Rationality: A Framework for Maintaining PSMs Number: 6 1. How RELEVANT is this paper to AI researchers? (mark one box) [x] Very relevant [ ] Moderately relevant [ ] Not relevant Please explain your rating: The authors of this paper use as solution to all problem types (reasoning forms) abduction with an BEST operators. They distinguish several types of BEST operators, however the idea is that solution of a problem can be find using abduction and an application of BEST operators. Applying an operator can be considered as a classification problem. This idea leads to using a maintaining method for classification for maintaining PSMs (because a PSM is only abduction and the right BEST operation.) They use the technique ripple-down-rules extended with functions. (I am not happy with this view of problem solving methods, but this is just only my opinion. The methods for specific tasks becomes so superfluous.) 2. How SIGNIFICANT is this paper? [ ] Very significant [x] Moderately significant [ ] Not significant Please explain your rating: 3. How ORIGINAL is this paper? [x] Very original [ ] Moderately original [ ] Not original Please explain your rating: I think it is a further development in the lines of the ECAI96 paper. New is the consideration of using a method for classification for maintenance of PSMs. The particular view of PSM (abduction+BEST operator) enables them to use a classification method. 4. Is this paper technically SOUND? [ ] Yes [x] Seems valid, but did not check completely [ ] Has minor errors [ ] Has major errors If there are errors, please identify them: 5. How well is this paper PRESENTED? Good Average Poor Overall organization [ ] [x] [ ] English [x] [ ] [ ] Readability [ ] [x] [ ] If there are presentational problems, please identify them: The first part of the paper is well written and very readable. The example section is for me at several places difficult to understand completely. 6. Further comments, advice or explanations (Please be specific and constructive, especially with respect to any negative judgements above. Point to the section(s) where an error occurs, cite omitted references, etc.) Use as much space as you need. Terminology: I would never called hypotheses as IN. In you terms given OUT, the IN will be computed. So for me IN and OUT are a little bit confusing. In your first example, the assumption set is {xUp,yUp, cDown, gUp, gDown} Could you explain why you consider also assumptions (e.g. yUp) which can be derived from other assumption (xUp) as assumptions? A nice categorization of BEST operators. You have to introduce V a little bit more. This because of the example of Reiter with the negation of abnormals. A question about section 2.4 Complexity: Could you give requirements on the theory, or IN, or OUT such that abduction becomes doable? In the example section: I do not understand why the knowledge base of fig. 6 is translated (fig.8) to using partialAND's. Is the idea that this _a_ way of a translation of _the_ way of translation. It is difficult to see, what the technique for maintaining single classification is and what the extensions are that are needed for the application to maintaining PSM. Possibly there are no extensions at all.