In this paper, we introduce a complete algorithm for computing the most specific hypothesis (MSH) in Inverse Entailment when the background knowledge is a set of definite clauses and the positive example is a ground atom having the same predicate symbol as that of the target predicate to be learned.
Muggleton showed that for any first order theory (background knowledge) B and a single clause (a positive example) E, the MSH can be computed by first computing all ground (positive and negative) literals which logically follow from B∧¬E and negating their conjunction. However, Yamamoto gave a counter example and indicated that Muggleton's proof contains error. Furukawa gave a sufficient condition to guarantee the above algorithm to compute the MSH. Yamamoto defined a class of problems where the algorithm computes the MSH. In this paper, we extend the MSH computation algorithm to ensure that it computes the MSH correctly under the condition described above.
[1]
Koichi Furukawa,et al.
On a Sufficient Condition for the Existence of Most Specific Hypothesis in Progol
,
1997,
ILP.
[2]
Alberto Maria Segre,et al.
Programs for Machine Learning
,
1994
.
[3]
Akihiro Yamamoto,et al.
Which Hypotheses Can Be Found with Inverse Entailment?
,
1997,
ILP.
[4]
Leo Breiman,et al.
Classification and Regression Trees
,
1984
.
[5]
J. Ross Quinlan,et al.
C4.5: Programs for Machine Learning
,
1992
.
[6]
Christopher M. Bishop,et al.
Classification and regression
,
1997
.
[7]
Leszek Pacholski,et al.
Undecidability of the Horn-clause implication problem
,
1992,
Proceedings., 33rd Annual Symposium on Foundations of Computer Science.