Modellieren, Lernen und Verarbeiten von Erfahrungswissen im Case-Based Reasoning auf der Grundlage präferenzbasierter Methoden - Präferenzbasiertes CBR

• Antragsteller: Professor Dr. Eyke Hüllermeier
• Fachliche Zuordnung: Bild- und Sprachverarbeitung, Computergraphik und Visualisierung, Human Computer Interaction, Ubiquitous und Wearable Computing
• Förderung: Förderung von 2010 bis 2016
• Projektkennung: Deutsche Forschungsgemeinschaft (DFG) - Projektnummer 170049638

• Erstellungsjahr: 2016

Zusammenfassung der Projektergebnisse

In this project, we have been working toward a methodological framework for case-based reasoning on the basis of formal concepts and methods for reasoning with preferences. Deviating from the common representation of experiences in terms of problem/solution tuples, preference-based CBR proceeds from weaker “chunks of information”, namely, preferences between competing solutions “contextualized” by problems: For a given problem X, a solution A is (likely to be) more preferred than another solution B. This type of information is often much easier to acquire in real applications. We developed a generic framework of preference-based CBR (called Pref-CBR), in which problem solving is realized as a search process that is guided by previous preference information: Having to solve a problem that resembles another problem already solved, the preferences observed for the latter are recalled and used to infer a presumably optimal adaptation of the current cadidate solution. Thus, in each step, the current best solution is compared with another, slightly modified/adapted solution, and the better one is retained. The general framework of Pref-CBR has been improved by methods for learning/adapting similarity measures and strategies for case base maintenance, which greatly increase its effectinveness. The type of application we have in mind is characterized by two important properties. First, since the evaluation of candidate solutions is expensive, only relatively few candidates can be considered in a problem solving episode before a selection is made. Second, qualitative feedback in the form of pairwise comparisons is much easier to acquire than a numerical assessment of individual candidate solutions. Several application domains complying with these assumptions, such as molecular docking and image processing, have been considered in the project and used as test beds for evaluating our methods. Overall, the original goals of the proposal have been achieved, and the project has been completed successfully. The success of the project is also documented by two best paper awards that have been given to papers on preference-based CBR, one at ICCBR 2011 and the other at ICCBR 2013. These awards not only confirm the quality of the research that has been conducted in the course of the project, but also the general interest of the CBR community in the topic of preference-based CBR.

Projektbezogene Publikationen (Auswahl)

• Preference-based CBR: First steps toward a methodological framework. In A. Ram and N. Wiratunga, editors, Proceedings ICCBR–2011, 19th International Conference on Case-Based Reasoning, number 6880 in LNAI, pages 77–91. Springer-Verlag, 2011. Best Paper Award
  E. Hüllermeier and P. Schlegel
  
• Preference-based CBR: A search-based problem solving framework. In S.J. Delany and S. Ontanon, editors, Proceedings ICCBR–2013, 21th International Conference on Case-Based Reasoning, number 7969 in LNAI, pages 1–14, Saratoga Springs, NY, USA, 2013. Springer. Best Paper Award
  A. Abdel-Aziz, W. Cheng, M. Strickert, and E. Hüllermeier
  
• Protein structure retrieval using preference-based cbr. In ICCBR Workshop on Case-based Reasoning in Health Sciences, Saratoga Springs, NY, USA, 2013
  A. Abdel-Aziz, M. Strickert, T. Fober, and E. Hüllermeier
  
• Learning solution similarity in preference-based CBR. In L. Lamontagne and E. Plaza, editors, Proceedings ICCBR–2014, 22nd International Conference on Case-Based Reasoning, number 8765 in LNAI, pages 17–31, Cork, Ireland, 2014. Springer
  A. Abdel-Aziz, M. Strickert, and E. Hüllermeier
  
• A CBR approach to the angry birds game. In Workshop Proceedings from ICCBR 2015, 23rd International Conference on Case-Based Reasoning, pages 68–77, Frankfurt am Main, Germany, 2015
  Adil Paul and Eyke Hüllermeier
  
• Case base maintenance in preference-based CBR. In Proceedings ICCBR–2015, 23rd International Conference on Case-Based Reasoning, number 9343 in LNAI, pages 1–14, Frankfurt am Main, Germany, 2015. Springer-Verlag
  A. Abdel-Aziz and E. Hüllermeier