Example-Based Learning and Reasoning

Dan Roth

For humans, looking at how concrete examples behave is
an intuitive way of deriving conclusions. The drawback 
of this method is that it does not necessarily give 
the correct results; most of our ``formal" theories 
of inference resort therefore to abstract knowledge
representations.

I will present two cases where examples can be used as
a ``good" knowledge representation.
First I will show that under certain conditions 
example-based deduction can be used to obtain a correct 
and complete inference procedure. Presented in the context 
of Boolean functions, this theory (reasoning with models) 
has applications in database theory and data mining.
The second instance will be the use of examples to represent
classifiers in very high-dimensional spaces. The main
focus here will be on the application to knowledge-intensive
inferences in the natural language domain.