Control of complex machines for meta-learning in computational intelligence

Recent years have revealed growing need for efficient meta-learning. For much longer time it has been known that there is no single adaptive algorithm, eligible to provide satisfactory (i.e. close to optimal) solutions for every kind of problem, however computing power facilitates practical applications of more and more sophisticated learning strategies and more and more thorough search in the space of candidate models. Because testing all possible models is not (and will never be) feasible, we need intelligent tools to combine human expert knowledge, the knowledge extracted by means of computational intelligence and different search strategies to disclose the nature of a problem and provide attractive models. We present some techniques, we have successfully used in our meta-learning approaches, describe the crucial ideas of our general architecture for meta-learning, and show some examples.

[1]  Saso Dzeroski,et al.  Combining Classifiers with Meta Decision Trees , 2003, Machine Learning.

[2]  Marek Grochowski,et al.  Comparison of Instances Seletion Algorithms I. Algorithms Survey , 2004, ICAISC.

[3]  Johannes Fürnkranz,et al.  An Evaluation of Landmarking Variants , 2001 .

[4]  Norbert Jankowski,et al.  Handwritten Digit, Recognition Road to Contest victory , 2007, 2007 IEEE Symposium on Computational Intelligence and Data Mining.

[5]  Salvatore J. Stolfo,et al.  JAM: Java Agents for Meta-Learning over Distributed Databases , 1997, KDD.

[6]  Melanie Hilario,et al.  Model selection via meta-learning: a comparative study , 2000, Proceedings 12th IEEE Internationals Conference on Tools with Artificial Intelligence. ICTAI 2000.

[7]  Hilan Bensusan,et al.  Meta-Learning by Landmarking Various Learning Algorithms , 2000, ICML.

[8]  Carlos Soares,et al.  Ranking Learning Algorithms: Using IBL and Meta-Learning on Accuracy and Time Results , 2003, Machine Learning.

[9]  Iain Paterson,et al.  Evaluation of Machine-Learning Algorithm Ranking Advisors , 2000 .

[10]  Carlos Soares,et al.  Ranking Classification Algorithms Based on Relevant Performance Information , 2000 .

[11]  Peter A. Flach,et al.  Improved Dataset Characterisation for Meta-learning , 2002, Discovery Science.

[12]  Norbert Jankowski,et al.  Heterogenous committees with competence analysis , 2005, Fifth International Conference on Hybrid Intelligent Systems (HIS'05).

[13]  Peter A. Flach,et al.  PKDD2000 workshop on Data Mining, Decision Support, Meta-learning and ILP : Forum for Practical Problem Presentation and Prospective Solutions , 2000 .

[14]  Philip K. Chan,et al.  Meta-learning in distributed data mining systems: Issues and approaches , 2007 .

[15]  Norbert Jankowski,et al.  Versatile and Efficient Meta-Learning Architecture: Knowledge Representation and Management in Computational Intelligence , 2007, 2007 IEEE Symposium on Computational Intelligence and Data Mining.

[16]  Włodzisław Duch,et al.  Committees of Undemocratic Competent Models , 2003 .