AKSW - Projects/DLLearner/Overview

2007-08-22 20:46:50

Comparison for /Projects / DL Learner / Overview of 2007-08-22 20:46:50 and 2008-08-05 15:44:45

Additions:

The following figure gives an overview of the “generate and test” approach used in DL-Learner and other inductive learning tools:

Deletions:

The following figure gives an overview of the DL-Learner architecture:

2007-08-22 20:10:58

Comparison for /Projects / DL Learner / Overview of 2007-08-22 20:10:58 and 2007-08-22 20:46:50

Additions:

See also the Wikipedia entry for ILP (Inductive Logic Programming). What DL-Learner considers is the the ILP problem applied to Descriptions Logics / OWL.

2007-08-22 19:56:09

Comparison for /Projects / DL Learner / Overview of 2007-08-22 19:56:09 and 2007-08-22 20:10:58

Additions:

Learning Problem

Given: background knowledge base, positive + negative examples (for a not yet defined concept)
Goal: a concept definition, such that when the definition is added to the background knowledge all positive examples follow and none of the negative examples follow

2007-08-22 19:00:33

Comparison for /Projects / DL Learner / Overview of 2007-08-22 19:00:33 and 2007-08-22 19:56:09

Additions:

The following figure gives an overview of the DL-Learner architecture:

Input for the learning algorithm are examples and background knowledge. DL-Learner itself is a framework for several learning algorithms like a refinement operator based algorithm and a genetic programming based algorithm (see publications for details). These algorithms use a generate and test approach. This means they generate concepts, which appear to be useful to them. They evaluate them using a concept quality measure (a heuristics). To measure the quality a Description Logic reasoner is required. The DL-Learner uses the so-called DIG interface to communicate with reasoners. This means that all DIG capable reasoners, i.e. alle popular reasoners, can be used. The concept generation procedure is repeated until a solution (or a sufficiently good concept) has been found.

Deletions:

ToDo: image of overall architecture (algorithm independant)

2007-08-22 18:44:25

2007-08-22 18:42:07

Comparison for /Projects / DL Learner / Overview of 2007-08-22 18:42:07 and 2007-08-22 18:44:25

Additions:

Deletions:

2007-08-22 18:41:38

Comparison for /Projects / DL Learner / Overview of 2007-08-22 18:41:38 and 2007-08-22 18:42:07

No differences.

2007-08-22 18:37:03

Comparison for /Projects / DL Learner / Overview of 2007-08-22 18:37:03 and 2007-08-22 18:41:38

Additions:

One of the simplest examples for concept learning is to learn the concept “father”. Assume the following background knowledge is given (Description Logic syntax):

In this case, an ontology learning system could learn the concept definition “Male AND EXISTS hasChild.TOP”, i.e. all objects, which are male and have at least one child.

Deletions:

One of the simplest examples for concept learning is to learn the concept “father”. Assume the following background knowledge is given:

// male and female are disjunct concepts
BOTTOM = (male AND female). 
// persons
male(markus).
male(stefan).
male(heinz).
male(bernd).
female(anna).
female(gabi).
female(michelle).
// children
hasChild(stefan,markus).
hasChild(markus,anna).
hasChild(bernd,gabi).
hasChild(anna,heinz).

Positive examples: stefan, markus, bernd

2007-08-22 16:08:33

Comparison for /Projects / DL Learner / Overview of 2007-08-22 16:08:33 and 2007-08-22 18:37:03

Additions:

Architecture

ToDo: image of overall architecture (algorithm independant)

Learning Example

One of the simplest examples for concept learning is to learn the concept “father”. Assume the following background knowledge is given:

// male and female are disjunct concepts
BOTTOM = (male AND female). 

// persons
male(markus).
male(stefan).
male(heinz).
male(bernd).
female(anna).
female(gabi).
female(michelle).

// children
hasChild(stefan,markus).
hasChild(markus,anna).
hasChild(bernd,gabi).
hasChild(anna,heinz).

Positive examples: stefan, markus, bernd