1
$\begingroup$

Deep learning seems to be the new cool thing in AI/machine learning and it works well in many domains, but I want to know- what are the specific application areas where deep learning is not the best approach and what is the reason for that?

Has some evaluation been made?

Is there a particular class of problems?

If so- what solutions are superior in solving that task?

$\endgroup$
1
$\begingroup$

Deep learning is generally not well suited for domains that have only very small training datasets. This includes fields like medicine where, for example, patient data can only be obtained through expensive and time consuming clinical trials. Clinical trials will often contain only several hundreds records.

By design, deep learning models have many parameters (e.g., millions), and like with any machine learning algorithm, you're likely to see overfitting when the number of parameters greatly exceeds the number of training records.

| improve this answer | |
$\endgroup$
  • $\begingroup$ so then what to do there? decision trees- or what? also--> how about SVM and kernels- do people still use this? how about decision logic? $\endgroup$ – smatthewenglish Jul 19 '16 at 8:34
  • $\begingroup$ When your dataset is only several hundred records it's often best to use linear models to reduce the risk of overfitting. An SVM or decision tree would also work, but you probably wouldn't want an ensemble of decision trees like a random forest or GBM. Yes, people still use SVMs. They work best on small to medium size datasets. $\endgroup$ – Ryan Zotti Jul 19 '16 at 11:35

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.