Showing papers by "Thomas G. Dietterich published in 1986"

Learning at the Knowledge Level

Abstract: When Newell introduced the concept of the knowledge level as a useful level of description for computer systems, he focused on the representation of knowledge. This paper applies the knowledge level notion to the problem of knowledge acquisition. Two interesting issues arise. First, some existing machine learning programs appear to be completely static when viewed at the knowledge level. These programs improve their performance without changing their ‘knowledge.’ Second, the behavior of some other machine learning programs cannot be predicted or described at the knowledge level. These programs take unjustified inductive leaps. The first programs are called symbol level learning (SLL) programss the second, nondeductive knowledge level learning (NKLL) programs. The paper analyzes both of these classes of learning programs and speculates on the possibility of developing coherent theories of each. A theory of symbol level learning is sketched, and some reasons are presented for believing that a theory of NKLL will be difficult to obtain.

Selecting appropriate representations for learning from examples

Abstract: The task of inductive learning from examples places constraints on the representation of training instances and concepts. These constraints are different from, and often incompatible with, the constraints placed on the representation by the performance task. This incompatibility explains why previous researchers have found it so difficult to construct good representations for inductive learning—they were trying to achieve a compromise between these two sets of constraints. To address this problem, we have developed a learning system that employs two different representations: one for learning and one for performance. The learning system accepts training instances in the "performance representation," converts them into a "learning representation" where they are inductively generalized, and then maps the learned concept back into the "performance representation." The advantages of this approach are (a) many fewer training instances are required to learn the concept, (b) the biases of the learning program are very simple, and (c) the learning system requires virtually no "vocabulary engineering" to learn concepts in a new domain.

Exploiting functional vocabularies to learn structural descriptions

Abstract: In the idea paper entitled “Learning Meaning,” Minsky [241] stresses the importance of maintaining different representations of knowledge, each suited to different tasks. For example, a system designed to recognize examples of cups on a table would do well to represent its knowledge as descriptions of observable features and structures. In contrast, a planning system employing cups to achieve goals would require a representation describing the purpose and function of cups. When we turn from the issue of employing a description of a cup to the task of learning such a description, it is not immediately obvious what vocabulary should be used. One approach might be to choose the vocabulary appropriate for the performance task (i.e., structural descriptions for recognition, functional descriptions for planning, etc.). This approach has been pursued, e.g., by Winston [396], Buchanan & Mitchell [46], Quinlan [282], and Minton [243]. In the case of Winston’s ARCH learner and Buchanan & Mitchell’s Meta-DENDRAL system, this approach worked well because good structural vocabularies were available. However, Quinlan and Minton confronted much more difficult problems in constructing structural vocabularies that concisely captured the desired game-playing concepts. Quinlan, for example, spent two man months developing the vocabulary for the concept of “lost-in-3-ply.”

The EG project: recent progress

Abstract: The long term goal of the EG project is to construct, implement, and evaluate a model of scientific inquiry. Figure 1 shows the model of scientific inquiry that we have developed to date.