Submitted by: Charles Taylor ************************************************************************** CALL FOR PAPERS MLNET FAMILIARIZATION WORKSHOP 26th April 1997 LEARNING IN DYNAMICALLY CHANGING DOMAINS: THEORY REVISION AND CONTEXT DEPENDENCE ISSUES. ************************************************************************** Up-to-date information will be kept at: http://www.amsta.leeds.ac.uk/statistics/ecml97/dyn.htm ********** BACKGROUND ********** In traditional Machine Learning the available examples (the training data) are usually used to learn a concept. In many practical situations in which the environment changes, this procedure ceases to work. Concerning the case of supervised learning, generally speaking, the application of a discrimination algorithm to classify new, unseen examples will be problematic if one of the following events occurs after the ``learning phase'': * The number of attributes changes * The number of attributes remains the same but the interpretation of the records of the datasets changes over time * A description of a concept (class) exists but there are additional databases relating to the given concept (class) that may modify (refine) the existing knowledge base In the case of supervised learning (classification or prediction), the extracted rules or dependencies can be no longer valid over time due to different changes which can occur. However, this statement is probably valid for unsupervised and reinforcement learning as well. ************** THE CHALLENGE: ************** There have been some attempts in the literature to address the problem of structural change in concepts and the dynamic aspects of data in Machine Learning. But most of the ML-algorithms can still not deal with this problem. Research in this direction has very important practical implications because structural change in concepts occurs often in the real-world domain. By considering this issue Machine Learning has a better chance of acceptance in industry and commerce. There are many contributions that Statisticians have (already) made to this field, but their communication is hampered by a different terminology, for example: ML Statistics -- ---------- Context Learning Parameter estimation in multivariate reg ression Dynamic Learning Structure Change (econometrics) Theory Revision, Knowledge Integration Hypothesis testing So, the question is: Have the Statisticians done everything; What is the challenge for ML? *************** RELEVANT TOPICS: *************** The ideal papers would be those which discuss (preferably two or more) learning aspects (from machine learning, Statistics and Neural Nets) in the following areas: * Structural change * CUSUM tests/ Quality control * Dynamic learning, dynamic models * Incremental learning, Sequential learning * Context learning * Theory revision * Knowledge Integration **************** AIMS AND PROGRAM **************** It is the aim of this workshop to : * Bring ML and statistics researchers together * Discuss the state of art of structural change in concepts. This discussion should not involve only symbolic ML but, specially, the statistical ML as well. Relevant contributions using Neural Networks are also welcome. * Discuss the direction for further research in the structural change in concepts bearing in mind that the main goal is solving real-world problems. The program (April 26) will include invited talks, presentations of accepted papers (both verbal and poster presentations). All of the contributions will be summarized by a member of the organizing committee in a talk. The contributions (including the invited talks) will be distributed as workshop notes. *********** ORGANIZERS *********** Gholamreza Nakhaeizadeh (Daimler-Benz, Germany) Charles Taylor (University of Leeds, UK) Ivan Bruha< (McMaster University, Canada) ********************* SUBMISSION OF PAPERS: ********************* Two kinds of submissions are solicited: full papers describing substantial completed research or applications, and poster papers reporting on work in progress. Submissions must be clearly marked as one of these two kinds. The program committee may decide to move accepted contributions from the full paper to the poster category. The size limit for submissions is 12 pages for full papers, 5 pages for poster papers (excluding title page and bibliography, but including all tables and figures). It is hoped that selected contributions will be subsequently published in an integrated volume. Submitted papers should preferably be formatted according to the LNAI guidelines (LaTeX style files are available at http://is.vse.cz/ecml97/styles.htm). Authors are encouraged to make their papers available in advance (by anonymous ftp or a URL site) so that wider discussion is possible. In future the above page will provide links to such papers. A separate title page must contain the title of the paper, the names and addresses of all authors, up to three keywords, and an abstract of max. 200 words. The full address, including phone, fax, and e-mail, must be given for the first author (or the contact person). The following items must be submitted by February 15, 1997: Either a camera-ready copy of the paper or a PS (uuencoded) file, together with an electronic version of the titlepage only (plain ASCII). Send submissions, enquiries, etc. to: Gholamreza Nakhaeizadeh (ECML-97) Daimler Benz AG Research and Technology Postfach 2360 D-89013 Ulm Germany e-mail: nakhaeizadeh@dbag.ulm.DaimlerBenz.COM Papers will be evaluated with respect to relevance, technical soundness, significance, originality, and clarity. Papers reporting on real-world applications will be evaluated according to special criteria. ************************************* REGISTRATION AND FURTHER INFORMATION: ************************************* The workshops will be open to anyone. Participants who are not members of MLnet pay a fee to cover the marginal costs of the workshop. The fee is yet to be determined. MLnet will pay the organisational costs for its members. MLnet will arrange travel bursaries for its members to take part in the workshops For information about paper submission and program, contact the program chair. F or information about local arrangements, registration forms, etc. contact the local organizers at actionm@cuni.cz **************** IMPORTANT DATES: **************** Submission deadline: 15 February 1996 Notification of acceptance: 8 March 1997 Camera ready copy: 1 April 1997 Workshop: 26 April 1997