Issue No. 04 - August (1996 vol. 11)
DOI Bookmark: http://doi.ieeecomputersociety.org/10.1109/64.511867
<p>This proposed learning system model lets engineers experiment more easily with alternative learning-tool configurations when developing knowledge-based applications.</p> <p>THE GROWING COMPLEXITY OF KNOWLEDGE-BASED applications makes it necessary to use automated or semiautomated techniques during knowledge acquisition. Knowledge engineering can also involve using various machine-learning systems, accepting different concept description languages or working with a variety of learning techniques that manage varied inputs such as example sets or background knowledge.</p> <p>For a specific task, knowledge engineers thus might need to experiment with different learning algorithms and their variations. In this context, they must </p> <p><li>1. select a machine-learning tool suited to the task, </li> <li>2. set the parameters that control the tool's behavior, and </li> <li>3. run the tool, analyze the results, and stop or return to steps 1 or 2.</li></p> <p>Running this loop might often be difficult and tedious. Selecting the appropriate tool might take time, and finding parameter settings well-suited to the problem's characteristics might also be time-consuming and difficult. Systems that provide a family of learning algorithms in a unified, rather than a single environment, can lessen this loop's complexity and duration.</p>
P. Albert and C. Rouveirol, "A Knowledge-Level Model of a Configurable Learning System," in IEEE Intelligent Systems, vol. 11, no. , pp. 50-58, 1996.