1
$\begingroup$

From this answer, stability is attributed to a learning algorithm

A stable learning algorithm is one for which the prediction does not change much when the training data is modified slightly.

At some other places, I read the phrase "stability of neural network model". I am not sure whether the stability of a learning algorithm and the stability of a model are the same or not. If same then

A stable model is one for which the prediction does not change much when the training data is modified slightly.

Is it true? If not, is there anything called stability of a model and is different from the stability of a learning algorithm?

Suppose I am training a neural network model with a gradient descent algorithm. For which one, do I need to attribute stability or instability? Is it to the neural network model or to the gradient descent training algorithm? Or it should be attributed to the combination of both?

$\endgroup$
3
  • $\begingroup$ is one for which the prediction does not change much when the training data is modified slightly. Is example back to do it again $\endgroup$ Aug 9 at 0:18
  • $\begingroup$ @Rachrachben Do you mean it applies to both model and training algorithm? $\endgroup$
    – hanugm
    Aug 15 at 23:50
  • $\begingroup$ Can you provide a source for the second part(stable model) as well? The answer you refer to cites a Wikipedia article which gives a list of stable learning algorithms. Neither gradient descent nor neural networks are on the list. $\endgroup$
    – serali
    Oct 9 at 12:13

Your Answer

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

Browse other questions tagged or ask your own question.