Information Technology Reference
In-Depth Information
The basic predicates are those that meet the need of operationality criterion.
The first column is all inoperable predicates. We will not make further
description in terms of constructing the dependent table. The followings are two
algorithms which completes the EBL.
Algorithm 9.3 Algorithm of generation of explanation tree
1. Inference backwardly from goal concept, extend the explanation tree
gradually, the tree is and/or tree.
2. Inverting resolution algorithm is called when a failed node is met.
3. Analyze the complete explanation tree and generalize to get a new description
of goal concept.
Algorithm 9.4 Algorithm of Inverting Resolution
1. If the current failed node F is operable, then backtracking; if F is inoperable, go to step
2.
2. The predicate symbol of node P corresponds to node F is Pred , check out if there is
other path for Pred based on dependent table, if there is no, go to step 4.
3. Check other paths or nodes of Pred in the dependent table whether their corresponding
operable predicates (basic predicates) can be satisfied by the training example.
If there is a path stratifies the training example, then choose this path,
explanation tree is completed, end the algorithm.
The corresponding operable predicates (basic property) of the rest of the path
or node of Pred in the dependent tree collide with training example or the
training example does not contain the path, go to step 4.
4. Infer all sub-nodes of node P except node F; go back to node F after dealing with all
the sub-nodes.
5. Apply inverting resolution to unused training properties to get a pseudo rule between
node F and the rest of properties, the creation of whole explanation tree is completed,
end of algorithm.
The algorithm described here can complete the EBL without perfect domain
theory, which is just short of very few rules. If the domain theory is so scatted
that can not form a complete knowledge structure, this algorithm will fail and
resort to ways of constructing knowledge base and tools, or to exert inductive
learning or analogy learning.
Last but not the least, although this algorithm can do the EBL, it lacks some
rules and thus introduces some inaccurate factors, the truth of the whole
explanation tree might be influenced and the identical truth can not be held as for
the explanation. If confidence can be added to the explanation tree, the
Search WWH ::




Custom Search