5
$\begingroup$

What are recommended methods for incorporating prior knowledge in machine learning tasks? By prior knowledge I am referring to information like which features are a priori considered more important or which samples should be weighted higher (e.g. due to higher data quality). I am looking for references that give a good overview of the available methods and also pointers to software implementations.

Improve this question
$\endgroup$
1
  • 1
    $\begingroup$ Well... When you have prior knowledge, Bayesian (machine learning) methods are always a good bet ;) $\endgroup$
    – air
    Commented Dec 23, 2015 at 11:36

1 Answer 1

Reset to default
2
$\begingroup$

I know this is 2 years old, but it might help someone at some point. Background knowledge comes in many forms, from prioritized features to whole Ontologies or other meta knowledge constructs. Dealing with this problem before, I have, for example, been able to represent my background knowledge as a graph (which it normally is anyways), and used its hierarchical properties to prioritize features when actually learning. As mentioned before, you can always resort to Bayesian methods, but the problem with them is the computational complexity, so be prepared to wait. A lot.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.