FOIDL is a descendant of FOIL differing from its predecessor in the
following three ways.
First, FOIDL is able to process intensionally defined background knowledge.
Second, the requirement to provide explicit
negative examples can be replaced by the assumption of output
completeness. The output completeness assumption requires
a mode declaration identifying input and output arguments of the target
predicate. Output completeness then states that for every unique
input pattern appearing in the training set, all correct
output patterns occur in the examples in the training set. Together
with the mode declaration, the positive examples then implicitly
determine the negative examples.
The third difference between FOIDL and FOIL is that FOIDL,
as FFOIL, induces decision lists.
Unlike FFOIL, FOIDL generates the clauses in the decision list in
reverse order,
that is, clauses learnt first appear at the end of the decision list.
As the covering algorithm tends to learn
more general clauses covering many positive examples first
the more general clauses are placed as default cases at
the end of the decision list.
References
R.J. Mooney and M.E. Califf.
Induction of first-order decision lists: Results on learning the
past tense of English verbs.
Journal of Artificial Intelligence Research, 3:1-24, 1995.