Information Technology Reference
In-Depth Information
Feature labels
+
−
−
+
−
Model
induction
Model
Feature values
−
+
Test instances
Figure 6.15
Active feature-value labeling: selecting feature values for association with
a certain class polarity.
Feature labels
+
−−
+
−
+
Model
induction
−
+
+
−
Dual-
trained
model
Feature values
−
+
Test instances
Figure 6.16
Active dual supervision: simultaneous acquisition of label information for
both feature values and instances.
been devised, practical, yet very simple solutions are easy to imagine. Consider
at the most basic extreme class assignment rules, where the presence of certain
feature values (or combinations of feature values) connects an example with a
particular class. A rudimentary “dual” model can also be constructed by simply
averaging the class predictions of a traditional supervised machine learning model
with a model based on feature-value/class relationships.
Search WWH ::
Custom Search