Call For Papers Machine Learning Special Issue on Applications of Machine Learning and the Knowledge Discovery Process Guest editors: Ronny Kohavi and Foster Provost With the explosion in size of business and scientific databases (VLDBs), the opportunities and pressure to mine the data and make novel discoveries have increased dramatically. For many problems, basic statistical summaries are not sufficient and there is a clear and recognized need for solutions involving a machine learning component. For example, modern businesses constantly seek to gain competitive advantage by tailoring actions to different customer segments and avoiding the trap of targeting the "average customer." This special issue of the journal Machine Learning will be dedicated to papers describing work in which machine learning technologies have been applied to solve significant real-world problems. In particular, it will focus on the application of Machine Learning technology, the simplifying assumptions that *cannot* be made in a real-world application, and the processes that are involved in going from the raw data to the final knowledge that decision makers seek. --------------- Scope --------------- High quality, original papers that address applications of machine learning technologies to significant, real-world problems are solicited. Authors are required to describe the papers' scientific contributions clearly. We actively solicit papers that address contributions not normally published in the machine learning literature, including (but not limited to): - - Applications of machine learning to real-world significant problems: successes, failures, limitations, and lessons learned. - - The knowledge discovery process: it is estimated that only 20% of the overall process is spent on running machine learning algorithms. What other things were done? What lessons were learned? Which parts of the overall process could be improved and perhaps automated? - - A novel solution to a non-trivial _class_ of applications. Such a paper would include: a description of the class of applications, detailing why they are different from previously solved applications; an "existence proof" on one member of the class; and a solid argument as to the generality of the solution for the class of applications. - - The identification of a number of assumptions normally made within machine learning research that cannot be made for this application, and a thorough description of how they were addressed (simultaneously) to engineer a working solution. - - A principled study of the relaxation of important assumptions necessary for the solution to the problem. - - Novel uses of existing algorithms and modifications to broaden their scope and overcome limitations and assumptions that hinder their use in real applications. Authors should address the following issues, when relevant: - - The real-world problem and its formulation as a machine learning task. How significant was the problem? - - Who are the users of the learned knowledge? Who "paid" for the work? How did their requirements constrain the knowledge discovery effort? How was success to be determined and was the expected payoff estimated in advance? - - The data selection process. Was a sampling process used or were all the data used? What were reasons for limiting the amount of data (resource constraints, simplicity of result, learning curve flattened early)? - - Data transformations. How were the raw data transformed into formats suitable for existing algorithms? What processes were required, why were certain transformations done, and were others tried? Examples: data cleaning, missing values, data transformations, aggregrations, group signatures/profiles, denormalization, and feature construction. - - Why were specific algorithms chosen, and which others were tried? - - What role did background knowledge play and how has it affected the process? - - What were the post-processing operations? How were the results explained to users (visualizations, dimensionality reduction, explanations, what-if scenarios, sensitivity analyses)? - - How did the results affect the target users? Was the mining done repeatedly, or was this a one-shot task? Was the learned knowledge or the learning system fielded and integrated in other processes? The contribution of each paper should be sufficient to lay the groundwork for future work on the topic. Future applied work can build on the initial solutions described; future academic work can explore the exposed assumptions in depth and provide more principled solutions. We encourage authors to suggest "challege problems" for future academic work. For example, can a useful "transformation language" be developed in a manner similar to the role relational calculus has served for relational databases? We hope that by providing the machine learning community with such a forum, we can stimulate the applied/academic cycle that is important for the healthy growth of a scientific field. --------------- Submission Requirements --------------- The Machine Learning journal is published by Kluwer Academic Publishers. Electronic submissions are ENCOURAGED; postscript copies may be e-mailed to mljapps@postofc.corp.sgi.com. Latex style file and related files available via anonymous ftp from ftp.std.com in Kluwer/styles/journals. For non-electronic submissions, send six (6) copies of the papers as indicated: Five (5) copies to: One (1) copy to: Mrs. Karen Cullen Ronny Kohavi MACHINE LEARNING Silicon Graphics, Inc, M/S 8U-876 Kluwer Academic Publishers 2011 N. Shoreline Boulevard 101 Philip Drive Mountain View, CA 94043 Assinippi Park Norwell, MA 02061 E-mail: ronnyk@sgi.com Tel: 415-933-3126 Manuscripts should be no longer than 8000-12000 words with full-page figures counting for 400 words. Shorter submissions, including technical notes are also solicited. The title page should include names, affiliations, and complete address including daytime telephone number and an electronic e-mail address of the contact author. Include a brief, one-paragraph abstract of 100-200 words and a list of keywords. Submissions must not have appeared in, nor be under consideration by, other journals. --------------- Important Dates -------------------- Submission deadline: 4 Mar 1997 Acceptance notification: 3 June 1997 ** Review criteria and further details will be sent in the future. Specific questions and clarifications should be sent to mljapps@postofc.corp.sgi.com