|
| Moderated by Stephen Muggleton. |
Hideo Bannai, Yoshinori Tamada, Osamu Maruyama, and Satoru MiyanoHypothesisCreator: Concepts for Accelerating the Computational Knowledge Discovery Process |
The
article
mentioned above has been submitted to the Electronic
Transactions on Artificial Intelligence, and the present page
contains the review discussion. Click here for
more
explanations and for the webpage of theauthors: Hideo Bannai, Yoshinori Tamada, Osamu Maruyama, and Satoru Miyano.
Overview of interactions
Q1. Anonymous Referee 1 (15.10):
Recommendation: This is an excellent paper and can be accepted for ETAI as it is. However, it seems to me that it would be useful if the authors add some words about previous papers in this field in order to compare the suggested method and known ones. Q2. Anonymous Referee 2 (15.10):
The paper is excellent and should be accepted, with minor revisions. These are detailed below. In the Introduction the abbreviation KDD is introduced. Please give the phrase of which it is an abbreviation. Probably it is "Knowledge Discovery from Databases". But how is the reader expected to know this? Does the first of the three bulleted points claim that the selection and/or development of appropriate algorithms is unimportant for computer-aided solution of discovery problems? If so, this is depressing news for those AI workers who currently specialize in the development of improved algorithms and principles of selection. Do the authors here suggest that these colleagues are wasting their time because numerous partially adequate methods exist already? If so, then compelling arguments should be given to support the implication that the severe gaps which exist in the capabilities of existing algorithms (such as inability to structure a problem by automatic discovery of intermediate concepts from raw data) can never be filled. The authors should consider whether the left hand column of Table 1 should be headed "Components ... " rather than "Elements ..." The latter term suggests that the items cannot be further decomposed. In the caption to Table 2 the MCC is defined as an algebraic ratio, from which the reader will be able to confirm that it indeed the bare prediction accuracy that is here measured. There is in this an implicit assumption that differential misclassification costs (i.e. that cost of a false positive is the same as cost of a false negative) can be safely made in this case, and that MCC can therefore be taken as an adequate measure of the quality of the classifier's performance. The author should draw this assumption to the reader's attention, and explain why in this application the penalties of predicting that a sequence contains a signal, when in fact it does not, are approximately equal to the penalties of failing to predict that a sequence contains a signal when in fact it does. In general summary, after attention to the need for minor revision, this paper deserves publication and makes an important contribution. |