Received: from ics.uci.edu by PARIS.ICS.UCI.EDU id aa16978; 13 Dec 91 15:07 PST Received: from ics.uci.edu by q2.ics.uci.edu id aa12148; 13 Dec 91 15:06 PST To: ML-LIST: ; Subject: Machine Learning List: Vol. 3 No. 23 Reply-to: ml@ics.UCI.EDU Date: Fri, 13 Dec 91 15:00:44 -0800 From: Michael Pazzani Message-ID: <9112131507.aa12148@q2.ics.uci.edu> Machine Learning List: Vol. 3 No. 23 Friday, Dec. 13, 1991 Contents: Knowledge Discovery In Databases Workshop Announcements at ML-92 The Machine Learning List is moderated. Contributions should be relevant to the scientific study of machine learning. Mail contributions to ml@ics.uci.edu. Mail requests to be added or deleted to ml-request@ics.uci.edu. Back issues may be FTP'd from ics.uci.edu in pub/ml-list/V/ or N.Z where X and N are the volume and number of the issue; ID: anonymous PASSWORD: ------------------------------ Date: Thu, 12 Dec 91 11:39:46 EST From: Gregory Piatetsky-Shapiro Subject: Book announcement K N O W L E D G E D I S C O V E R Y I N D A T A B A S E S Edited by Gregory Piatetsky-Shapiro and William J. Frawley AAAI Press / The MIT Press 525 pages. paperback. Addressed to computer scientists, MIS professionals, and those interested in machine learning and databases, this collection is the first to bring together leading edge research in the exciting field of discovery in databases. Its thirty chapters span data-driven, knowledge-based, and integrated approaches, dealing with the discovery of quantitative and qualitative laws, data summarization, methodology, and application issues. From the foreword by J.R. Quinlan: "Knowledge Discovery in Databases is a pretty ambitious title ... but in the best sense of capturing the essence of something that is both achievable and worth attaining. The 1990s should see the widespread exploitation of knowledge discovery as an aid to assembling knowledge bases." Contents ------------------------ Foreword, J. R. Quinlan 1. Knowledge Discovery in Databases - An Overview. W. Frawley, G. Piatetsky-Shapiro, C. Matheus. -------- Part I. Discovery of Quantitative Laws. 2. Interactive Mining of Regularities in Databases. J. Zytkow, J. Baker 3. Discovering Functional Relationships from Observational Data. Y.-H. Wu, S. Wang 4. Minimal-length Encoding and Inductive Inference. E. Pednault 5. On Evaluation of Domain-Independent Scientific Function-Finding Systems. C. Schaffer -------- Part II Discovery of Qualitative Laws. post6. A Statistical Technique for Extracting Classificatory Knowledge from Databases. K.C.C. Chan, A.K.C. Wong 7. Information Discovery through Hierarchical Maximum Entropy Discretization and Synthesis. D.K.Y. Chiu, A.K.C. Wong, B. Cheung 8. Learning Useful Rules From Inconclusive Data. R. Uthurusamy, U. Fayyad, S. Spangler 9. Rule Induction using Information Theory. P. Smyth, R. Goodman 10. Incremental Discovery of Rules and Structure by Hierarchical and Parallel Clustering. J. Hong, C. Mao 11. The Discovery, Analysis, and Representation of Data Dependencies in Databases. W. Ziarko --------- Part III - Using Knowledge in Discovery. 12. Attribute-Oriented Induction in Relational Databases. Y. Cai, N. Cercone, J. Han 13. Discovery, Analysis, and Presentation of Strong Rules. G. Piatetsky-Shapiro 14. Integration of Heuristic and Bayesian Approaches In a Pattern Classification System. Q. Wu, P. Suetens, A. Oosterlink 15. Using Functions to Encode Domain and Contextual Knowledge in Statistical Induction. W. Frawley 16. Integrated Learning in a Real Domain. F. Bergadano, A. Giordana, L. Saitta, F. Brancadori, D. De Marchi 17. Induction of Decision Trees from Complex Structured Data. M. Manago, Y. Kodratoff ---------- Part IV - Data Summarization 18. Summary Data Estimation using Decision trees. M.C. Chen, L. McNamee 19. A Support System for Interpreting Statistical Data. P. Hoschka, W. Kloesgen 20. On Linguistic Summaries of Data. R. Yager ---------- Part V - Domain-Specific Discovery Methods 21. Extracting Reaction Information from Chemical Databases. C.-S. Ai, P. Blower, R. Ledwith 22. Automated Knowledge Generation From a CAD Database. A. Gonzalez, H. Myler, M. Towhidnejad, F. McKenzie, R. Kladke 23. Justification-Based Refinement of Expert Knowledge. J. Schlimmer, T. Mitchell, J. McDermott 24. Rule Discovery for Query Optimization. M. Siegel, E. Sciore, S. Salveter ---------- Part VI - Integrated and Multi-Paradigm Systems 25. Unsupervised Discovery in an Operations Control Setting. B. Silverman, M. Hieb, T. Mezher 26. Mining for Knowledge in Databases: Goals and General Description of the INLEN System. K. Kaufman, R. Michalski, L. Kerschberg ---------- Part VII - Methodology and Application Issues 27. Automating the Discovery of Causal Relationships in a Medical Records Database: The POSCH AI project. J. Long, E. Irani, J. Slagle 28. Discovery of Medical Diagnostic Information: An Overview of Methods and Results. M. McLeish, P. Yao, M. Garg, T. Stirtzinger 29. The Trade-Off Between Knowledge And Data in Knowledge Acquisition. B. Gaines 30. Knowledge Discovery as a Threat to Database Security. D. O'Leary ---------------------------------- To order contact MIT Press 55 Hayward Street Cambridge, MA 02142, USA or call toll-free (from USA) 1-800-356-0343 ISBN number 0-262-66070-9, Book code PIAKP Price: $37.50 + 2.75 postage. International postage $4.75 - surface, $8.50 airmail ------------------------------ Date: Fri, 13 Dec 91 14:22:07 EST From: gordon@aic.nrl.navy.MIL Subject: workshop announcements *************************************************************************** CALL FOR PAPERS Informal Workshop on ``Biases in Inductive Learning" To be held after ML-92 Saturday, July 4, 1992 Aberdeen, Scotland All aspects of an inductive learning system can bias the learn- ing process. Researchers to date have studied various biases in inductive learning such as algorithms, representations, background knowledge, and instance orders. The focus of this workshop is not to examine these biases in isolation. Instead, this workshop will examine how these biases influence each other and how they influence learning performance. For example, how can active selection of instances in concept learning influence PAC convergence? How might a domain theory affect an inductive learning algorithm? How does the choice of representational bias in a learner influence its algo- rithmic bias and vice versa? The purpose of this workshop is to draw researchers from diverse areas to discuss the issue of biases in inductive learning. The workshop topic is a unifying theme for researchers working in the areas of reformulation, constructive induction, inverse resolu- tion, PAC learning, EBL-SBL learning, and other areas. This workshop does not encourage papers describing system comparisons. Instead, the workshop encourages papers on the following topics: - Empirical and analytical studies comparing different biases in inductive learning and their quantitative and qualitative influ- ence on each other or on learning performance - Studies of methods for dynamically adjusting biases, with a focus on the impact of these adjustments on other biases and on learning performance - Analyses of why certain biases are more suitable for particular applications of inductive learning - Issues that arise when integrating new biases into an existing inductive learning system - Theory of inductive bias Please send 4 hard copies of a paper (10-15 double-spaced pages, ML-92 format) or (if you do not wish to present a paper) a descrip- tion of your current research to: Diana Gordon Naval Research Laboratory, Code 5510 4555 Overlook Ave. S.W. Washington, D.C. 20375-5000 USA Electronic and FAX submissions will not be accepted. If you have any questions about the workshop, please send email to Diana Gordon at gordon@aic.nrl.navy.mil or call 202-767-2686. Important Dates: March 12 - Papers and research descriptions due May 1 - Acceptance notification June 1 - Final version of papers due Program Committee: Diana Gordon, Naval Research Laboratory Dennis Kibler, University of California at Irvine Larry Rendell, University of Illinois Jude Shavlik, University of Wisconsin William Spears, Naval Research Laboratory Devika Subramanian, Cornell University Paul Vitanyi, CWI and University of Amsterdam ************************************************************************ CALL FOR PARTICIPATION Informal Workshop on ``Knowledge Compilation and Speedup Learning'' To be held after ML-92 Saturday, July 4, 1992 Aberdeen, Scotland Description of Workshop ``Knowledge Compilation'' is the problem of converting a declarative specification or a domain theory to an efficient executable program. ``Speedup Learning'' is the problem of improving, or speeding up, a slow problem solver. These two tasks are closely related since a declarative specification or a domain theory can be viewed as a slow problem solver. The approaches to knowledge compilation or speedup learning include explanation-based learning, empirical learning, and partial evaluation, among others. This workshop aims to bring together researchers in all these areas with the goal of achieving a better understanding of knowledge compilation and speedup learning. This workshop is also a sequel to the ``Knowledge Compilation'' workshop which was organized by Jim Bennett, Tom Dietterich, and Jack Mostow in 1986. There have been a number of results -- both positive and negative -- in speedup learning in the past few years. For example, while there were some positive results in Explanation-Based Learning (EBL) (e.g., Minton), the ``utility problem,'' or the exorbitant computational cost of using the learned knowledge, proved to be a significant obstacle to further progress (e.g., Minton, Tambe and Rosenbloom). On the other hand, there have also emerged some interesting relationships between partial evaluation and EBL (e.g., van Harmelen and Bundy), and between EBL and empirical learning (e.g., Yoo and Fisher). This appears to be a good time to consolidate what we know from all these areas and define the next round of research problems and issues. Specifically, we are interested in questions like the following: -- what does it mean to compile knowledge or speedup a problem solver? -- what are the factors that influence the speedup? -- what are the conditions under which successful speedup can occur? -- how do examples help in obtaining speedup? -- what are some of the methods that can achieve compilation/speedup? -- are different speedup/compilation methods distinguishable? -- what are the conditions of success for various methods? -- how does the learning protocol effect the speedup? -- how does speedup learning relate to concept learning? -- how does the language of knowledge representation influence speedup learning? -- how to scale up speedup learning methods to real world tasks? -- what are some of the promising research directions in this area? This is not an exhaustive list but a sample of questions that we are interested in. We specifically want to emphasize informal discussions and exchange of ideas, rather than polished research results. In this spirit, we welcome position papers on the research directions as well as rational reconstructions of previous work. Especially encouraged are papers that elucidate the relationships between different approaches to speedup learning/knowledge compilation (e.g., explanation-based learning and empirical learning). If you wish to do a presentation at the workshop, please email an extended abstract (of a maximum of 1000 words) to tadepalli@cs.orst.edu by March 20, 1992. The file should be in ascii or postscript only. If email is impossible, then please send 4 copies of the abstract to: Prasad Tadepalli Department of Computer Science 303 Dearborn Hall Oregon State University Corvallis, OR 97331-3202 If you do not wish to present a paper but want to attend the workshop, please send a one page summary of your relevant research and publications to the same address by the above date. Important Dates: March 20 - Abstracts and research descriptions due May 1 - Acceptance notification June 1 - Final version of papers due Program Committee: Oren Etzioni, University of Washington Doug Fisher, Vanderbilt University Nick Flann, Utah State University Steve Minton, NASA Ames Research Center Armand Prieditis, University of California Devika Subramanian, Cornell University Prasad Tadepalli, Oregon State University Frank van Harmelen, University of Amsterdam ************************************************************************ CALL FOR PAPERS Informal Workshop on ``Computational Architectures for Supporting Machine Learning and Knowledge Acquisition'' To be held after ML-92 Saturday, July 4, 1992 Aberdeen, Scotland The synthesis and refinement of a knowledge base has been the focus of research in both the knowledge acquisition and machine learning communities. Both groups, working independently, have produced with varying degrees of success, a number of systems for producing or refining a knowledge base. Knowledge acquisition (and apprentice) systems have considered these problems by focusing on how a human expert can be used to identify knowledge or to identify errors and suggest modifications. Machine learning uses examples of concepts and/or a domain theory to induce or modify a knowledge base. Although both communities rely on very different sources of knowledge, both face the key issues of defining what needs to be learned and techniques that are useful for learning this knowledge. Identifying what is to be learned depends in part on the description of the problem solving goals for which the knowledge base might be used. There are many ways to describe these goals, and the proliferation of many different kinds of knowledge-based system descriptions (e.g. rule bases, theorem provers, problem spaces, task specific architectures) is evidence of this. The difference between these descriptions comes from the models of the problem solving process the knowledge base is to support. For example, many knowledge-based systems for problems such as diagnosis are described as collections of rules. Another way to describe the same system is as a collection of tasks that decompose into subtasks. Regardless of the type of description of a knowledge base that is used, each one defines a particular system architecture whose organization has great impact on what is to be learned. The architecture essentially defines the knowledge that is to be learned or modified. Just as the architecture defines the types of knowledge that are needed and affects the efficiency and effectiveness of the knowledge base's performance, the architecture also has great impact upon the efficacy of the learning process. Usually the number of concepts that can be learned or modifications that can be made is very large, and for a learning system to operate efficiently, it must be able to constrain the space of alternatives to those most likely to be useful. This is the role of bias in learning systems -- to guide the learning toward a subset of possible concepts, which improve performance. The system architecture is an excellent guide to the selection of bias. Unfortunately, there has been little interaction between the knowledge acquisition, machine learning, and problem solving communities. Most of the discussion in these areas has also focused on techniques for synthesizing or modifying knowledge instead of considering how the architecture constrains what is to be learned. In this workshop, we will investigate the relationship between architectures and constraints on the learning process by considering how a number of architectures support the identification of what needs to be learned. In particular, we invite researchers who have implemented knowledge acquisition and machine learning systems based upon such architectures to present their results, so that in a comparative, cooperative session the entire community can profit from their successes and learn from their limitations. The workshop will consist of a series of presentations and panel discussions by researchers that describe the impact of the knowledge base's architecture on their learning experiments. The discussion will focus on the types of architectures and their role in defining the learning space. The goal of the workshop is to try to identify characteristics of an architecture that provide power for learning. By having an open forum where examples of how each architecture affected the learning process, the workshop will provide a useful way to investigate this issue. In this workshop, we invite submissions that discuss how an architecture they used defined what is to be learned. Points a submission might address include: - a description of the architecture of the knowledge base (e.g. rule base, predicates, hierarchy of tasks, problem spaces, etc.) using a vocabulary of domain independent terms that are used by the learning system - a description of the kinds of knowledge the learning system gathered in knowledge base synthesis or the kinds of errors that are identified in knowledge base refinement, more specifically, -- in knowledge base synthesis: --- how did the architecture focus the knowledge gathering process (what is the space of knowledge to be learned and how does the architecture limit this space), --- what is the vocabulary of the questions that are asked of the human expert defined by the architecture, --- how does the architecture affect the user interaction with the learning system --- how directly does the architecture define what is to be learned (how much interpretation does the human expert or system need to decide what is to be learned) --- what kinds of examples are used and how are they described -- in knowledge base refinement: --- what are the kinds of errors that are defined by the architecture --- how many possible errors are typically considered in a case --- what kinds of knowledge are needed to determine which error(s) occurred --- what kinds of knowledge are required to modify the knowledge base - a discussion of the results of their learning system -- what leverage did the architecture provide on their learning problem -- what limitations or costs did the architecture cause in learning Submissions should be 5-7 pages in length in the format for the ML conference, and they should be accompanied by a 1-2 page description of your current research. People wishing to attend the workshop but not present a paper *need only* send a research description. Please send 5 copies of your submisssion and/or research description to: Michael Weintraub GTE Laboratories 40 Sylvan Road Waltham, MA 02254 USA Important Dates: March 15 - Papers and research descriptions due May 1 - Acceptance notification June 1 - Final version of papers due Program Committee: Dean Allemang, Istituto Dalle Molle di Studi sull'Intelligenza Artificiale, Lugano (dean@idsia.ch) Ray Bareiss, Northwestern University (bareiss@ils.nwu.edu) Susan Craw, Robert Gordon Institute of Technology, Aberdeen (smc@csd.abdn.ac.uk) Henrik Eriksson, Linkoping University (eriksson@sumex-aim.stanford.edu) Ashok Goel, Georgia Institute of Technology (goel@cc.gatech.edu) Tom Gruber, Stanford University (gruber@sumex-aim.stanford.edu) Mark Musen, Stanford University (musen@sumex-aim.stanford.edu) Maarten van Someren, University of Amsterdam, (maarten@swi.psy.uva.nl) Michael Weintraub (Chair), GTE Laboratories (maw2@gte.com) ********************************************************************* CALL FOR PAPERS Informal Workshop on ``Integrated Learning in Real-world Domains'' To be held after ML-92 Saturday, July 4, 1992 Aberdeen, Scotland Experience has shown that many learning paradigms fail to scale up to real-world problems. One response to these failures has been to construct systems which use multiple learning paradigms. Thus, if one paradigm succeeds at the failure points of others, the effectiveness of the overall system will be enhanced (i.e., coverage is gained). Consequently, integrated techniques have become widespread over the last few years. Such systems can be viewed on a spectrum ranging from "tool-box" approaches to tightly-integrated systems. In a tool-box approach, a number of different learning paradigms are packaged together. The user, faced with a particular problem, decides which paradigm to use in this instance, or how to combine paradigms to solve the problem jointly. Examples of "tool-box" systems include Michalski's MTL, EMERALD, and INLEN. In tightly-integrated systems, several machine learning paradigms are combined to produce a new technique. Examples of tightly-integrated systems include Danyluk's Gemini, Pazzani's OCCAM, Cohen's A-EBL and GRENDEL, Pazzani, Brunk \& Silverstien's FOCL, Shavlik \& Towell's ANN-EBL, Mooney \& Ourston's EITHER, and Bergadano, Giordana \& Saitta's ML-SMART. Between these two ends of the spectrum are loosely-coupled systems where the different learning algorithms all attempt more or less independently to solve either all or part of the problem, perhaps cooperating somewhat. GTE's ILS is an example of this type of system. This workshop is intended to bring together researchers combining empirical, knowledge-intensive, neural net (and other connectionist approaches), genetic, case-based, statistical (classical, Bayesian, and non-parametric approaches) or other learning techniques to solve real-world problems. Special consideration will be given to systems which combine techniques from a wide variety of fields. This workshop will encourage papers on the following topics: - a case study detailing useful information learned from the application of an integrated technique to a real-world domain, - results on novel combinations of paradigms, or - comparative analysis (either empirical or theoretical) of relative performance of individual learning paradigms and integrated techniques Additionally, each paper should answer the following questions: - What is "real" about your domain? - Do you have large quantities of data? is it reliable? - Do you have a source of expert domain knowledge? What is it? Is it reliable? - Is it important that anyone ``understand'' the results of your learning system? - Is it sufficient to show improvement of a performance system empirically? - Need the learning system justify itself? - Need it work cooperatively with an expert? - Why is a single learning paradigm inadequate? - When working in a real domain, does the combination of paradigms applied become so tailored to the problem that it does not apply more generally? Please send 4 copies of a paper (max. 10 pages, ML-92 format) or (if you do not wish to present a paper) a description of your current research to the workshop chairperson: Patricia Riddle Boeing Computer Services P.O. Box 24346, MS 7L-64 Seattle, Washington USA 98124-0346 Important Dates: March 15 - Papers and research descriptions due May 15 - Acceptance notification May 29 - Final version of papers due Program Committee: Bernard Silver, GTE Laboratories Andrea Danyluk, NYNEX Science and Technology Michael Pazzani, UC Irvine Steve Chien, Jet Propulsion Laboratory Lorenza Saitta, Universita' di Torino Yves Kodratoff ************************************************************************* CALL FOR PAPERS Workshop on ``Machine Discovery'' To be held after ML-92 Saturday, July 4, 1992 Aberdeen, Scotland The number of researchers working on machine discovery (scientific discovery, knowledge discovery in databases, automation of data analysis, and other areas) is currently greater than one hundred and growing. A substantial number of new projects are being developed and plenty of interesting results can be shared. Discovery researchers constitute an important group within machine learning, driven by specific interests, applications, and evaluation mechanisms. Machine Discovery Workshop will be the place for them to gather and discuss the specialized topics of the discovery research. Several overlapping communities will have a chance to meet, including, among others, those who work on scientific discovery, those who focus on knowledge discovery in data bases, and those dealing with data analysis and discovery of data dependencies. The program will consist of paper presentations, panel discussion, and demonstration of machine discovery systems. All accepted papers will be published in the proceedings. Some of those will be presented during the poster session. Topics of interest include, but are not limited to: - Scientific discovery: empirical discovery, data driven reasoning, theory revision, discovery of quantitative laws, discovery of hidden structure, experiment design and planning, theory driven reasoning, domain applications and cognitive models; - Discovery in databases: discovery of regularities and concepts, discovery of data dependencies, discovery of causal relations, use of domain knowledge; - Automated data analysis: data and concept classification, combining search with statistics, search for empirical equations; - other: integrated and multiparadigm systems, exploration of environment, evaluation mechanisms, domain-specific discovery methods, mathematical discovery and discovery in abstract spaces. REQUIREMENTS FOR SUBMISSION: The papers must not exceed 10 single spaced pages, not counting bibliography, but including abstract of 180-220 words. Submissions in the category of demos: 3 page description of the system plus a sample run of the system (up to 3 pages; commented), plus answers to the questionnaire, mailed on request. IMPORTANT DATES: Submissions (on paper; in 4 copies) must arrive by March 31 to Jan Zytkow Computer Science Department Wichita State University Wichita, KS 67208 email: zytkow@wsuiar.wsu.ukans.edu phone: 316-689-3178 Notifications of acceptance will be sent on April 29 (provide your e-mail address, if possible). Camera-ready copies must arrive by June 1. PROGRAM COMMITTEE (and ORGANIZING COMMITTEE) Peter Edwards University of Aberdeen, United Kingdom Ken Haase MIT, USA Jiawei Han Simon Fraser University, Canada Peter Karp SRI International, USA Willi Klosgen German National Research Center for CS Yves Kodratoff Universite Paris-Sud, France Deon Oosthuizen University of Pretoria, South Africa Paul O'Rorke Univ.of California, Irvine, USA Gregory Piatetsky-Shapiro GTE Laboratories, USA Armand Prieditis Univ.of California, Davis, USA Cullen Schaffer CUNY/Hunter College, USA Derek Sleeman University of Aberdeen, United Kingdom Raul Valdes-Perez Carnegie-Mellon, USA Robert Zembowicz Wichita State Univ., USA Wojciech Ziarko Univ. of Regina, Canada Jan Zytkow Wichita State Univ. USA (workshop chair) ------------------------------ END of ML-LIST 3.23