Learning Descriptors Through Generalization And Discovery

Qingming He, Zhijun He, Ruizhad Yu · 2005

Descriptor is an elementary concept for describing objects or situations. We distinguish two kinds of descriptor: elementary descriptor and constructive descriptor. The constructive descriptor is defined by other descriptors. This paper presents a constructive descriptor learning system --- LEDES. LEDES integrates the facilities in learning from examples and learning from discovery to learn descriptor's definition, and keeps it in a hierarchy for subsequent learning. LEDES is an incremental and active learning system, i.e. its learning is multistage, and it can construct examples to ask trainer in ambiguous situations. Concept learning is a peculiar descriptor learning.

Read the paper · More papers on PaperTik