Multiple Intelligent Agents for Pattern Recognition Tasks
Louis Vuurpijl & Lambert Schomaker

The preliminary working title of this project is MAPR, an acronym for Multiple Agents for Pattern Recognition. In this postdoc project, the multiple-agent paradigm (MA) is investigated for pattern recognition tasks. In particular, the attention is focused at the integration of heterogeneous information (i.e., of different type and granularity) from multiple sources (e.g., raw input signal, contextual information, knowledge base of previous samples). In MAPR, the MA concept is being used as an alternative to the traditional statistical and algorithmic classifier approaches.

This document is always in progress.

An intelligent agent is an autonomous software component with goals (what can you achieve?), beliefs (how can you achieve?) and confidences (how certain are you?). Current MA technology comprises:

• frameworks for developing distributed multiple agents which communicate according to a specified protocol and negotiate following a specific negotiation scheme,
• formal negotiation models originating from game theory, multi-sensor fusion or research in economics,
• machine learning capabilities like case based reasoning and genetic programming,
• knowledge representation systems (like http://ontolingua.nici.kun.nl).

The MAPR project intends to integrate findings of each of these research areas, and develop applications in the areas of online text recognition and image retrieval on the Web. We currently have developed a MA framework running on Internet and implementing a digit-recognition system [1]. Another application, combining information generated by a shape-based classifier and knowledge provided by a language parser, is presented in [2].

A related PhD project Heterogeneous information integration: Theory and application of the multiple-agent paradigm in pattern recognition problems is started in September 1998.

1. L. Vuurpijl, L. Schomaker, Multiple-Agent Architectures for the Classification of Handwritten Text, IWFHR6, International Workshop on Frontiers of Handwriting Recognition, pp. 335--346, Seoul, August 1998

2. L. Schomaker, E. Hoenkamp, M. Mayberry, Toward collaborative agents for automatic on-line handwriting recognition, Third European Workshop on Handwriting Analysis and Recognition, pp. 13/1--13/6, Brussels, July 1998

3. L.G. Vuurpijl and L.R.B. Schomaker. A framework for using multiple classifiers in a multiple-agent architecture. Proceedings of the third IEE European workshop on Handwriting analysis and recognition. Pages 8/1--8/6, July 14, 1998.
   [(.ps.gz) 60067 bytes], [(.pdf)

4. Louis Vuurpijl and Lambert Schomaker. Two-stage character classification: a combined approach of clustering and support vector classifiers. Proceedings of IWFHR7, the seventh International Workshop on Frontiers in Handwriting Recognition. Pages 423-432, Amsterdam, The Netherlands, September 2000.
   [(.ps.gz) 105244 bytes], [(.pdf)

5. Fusi Wang, Louis Vuurpijl and Lambert Schomaker. Support Vector Machines for the Classification of Western Handwritten Capitals. Proceedings of IWFHR7, the seventh International Workshop on Frontiers in Handwriting Recognition. Pages 167-176, Amsterdam, The Netherlands, September 2000.
   [(.ps.gz) 114249 bytes], [(.pdf)

1. MAPR background
   This link is kept under passwd, because of the exciting, but still unpublished ideas
2. MAPR applications
3. MAPR publications
4. Agent Links