The Machine Learning Toolbox contains a set of ten Machine Learning algorithms, integrated with a common interface and common knowledge representation language. An essential component of the Toolbox is the Consultant, a knowledge-based system that advises novice users about which algorithm they could use for a particular application. We show how the Consultant’s architecture evolved, through its successive implementations, from a rigid rule-based expert system to a flexible information browsing system supporting user experimentation. In particular, we show how a task description can be elicited from the user in three different modes and exploited by several functions to provide advice and explanations at various levels of detail. The system’s output also increased in sophistication: initially limited to the recommendation of a suitable algorithm, it now includes detailed information about the algorithm and its usage, and will be extended to help the user interpret and improve the results of learning.
[1]
Michael Hucka,et al.
An analysis of soar as an integrated architecture
,
1991,
SGAR.
[2]
Nigel Shadbolt,et al.
Using Generalised Directive Models in Knowledge Acquisition
,
1992,
EKAW.
[3]
Marinus J. Bouwman.
Intelligent tutoring systems: D. SLEEMAN and J.S. BROWN (Eds.) Computers and People Series, Academic Press, London, 1982, xi + 345 pages, £19.20
,
1983
.
[4]
E. Demonchaux,et al.
Data analysis, data bases and expert systems: the common interface.
,
1988
.
[5]
E. Morales.
The Machine Learning Toolkit Database
,
1990
.