1
$\begingroup$

From the Thomas Dietrich's article Ensemble Methods in Machine Learning, this 3 motivations can be concluded as: Statistical, Computational and Representational.

Could anyone concluded each motivation with a short word explanation(description)

$\endgroup$
  • $\begingroup$ sorry, I mean short description for each motivation, and I just modify it. $\endgroup$ – HungryBird May 10 '18 at 20:00
  • $\begingroup$ sorry i was joking, but such a question may be too broad to answer. my feeling is that the reason we have ensemble learning is because we want to build a more accurate predictor. So more than one predictors are needed... $\endgroup$ – Haitao Du May 10 '18 at 20:05
  • $\begingroup$ you are right, this is exactly why we need ensemble learning, but what I want to get from this post is something like this: statistical, what is statistical problem, how do ensemble solve it. $\endgroup$ – HungryBird May 10 '18 at 20:11
2
$\begingroup$

Here are some points that I get from Thomas Dietrich's article Ensemble Methods in Machine Learning that might help you

Statistical

The statistical problem can be found when the amount of training data is too small than the size of search space that causes the learning algorithm give the same accuracy on the training data although it is different hypotheses. By using ensemble methods, the algorithm can average the accurate hypotheses, reduce the risk of choosing the wrong classifier and also can find a good approximation to the true hypothesis.

Computational

The computational problem arise when training data is enough but learning algorithm is still stuck in local optima. So it's still very difficult computationally for the learning algorithm to find the best hypothesis. An ensemble constructed by running the local search from many different starting points may provide a better approximation to the true unknown function than any of the individual classifiers.

Representational

In most applications of machine learning, the true function can be represented by forming weighted sums of hypotheses. This may be possible to expand the space of representable functions. So, we must consider the chosen search space to be the effective space of hypotheses searched by the learning algorithm for a given training dataset.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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