Modeling Identification Strategies Using the MCC Methodology

Identification is the task of attributing an object to a class, generally called classification of this object. The goal of this work is to identify, represent and finally acquire identification strategies. In this purpose, we have modeled the identification task on the basis of two existing algorithms: XPER and MAKEY that use the same type of knowledge base as input. We have identified a common problem solving method for identification called "step by step identification". Modelling this particular problem solving method in the MCC methodology, a methodology specifically adapted to the representation of control knowledge, allows to represent clearly identification strategies as control knowledge specific to the method. We present in the paper the content of identification strategies and show the need for such strategies in particular domains such as Systematics and Medecine. Our problematic is in the center of the two research fields of Knowledge Acquisition (KA) and Machine Learning (ML). We try to improve a ML tool (MAKEY) by modeling its behavior at the knowledge level, so as to explicit its control. This done, we can represent declaratively the control knowledge needed by the tool, and use it to make the tool more adaptable to real problems. In addition, this clear representation of control knowledge will render us capable of acquiring the control knowledge from experts.