I'm working on a time-series (comm signals) classification problem that has 25 possible classes. I have one fairly deep CNN that can perform fairly well on this problem, but I'm hoping to make better. So I'm wondering if there's merit in using a few CNN's to solve the same classification problem instead.

Specifically, the 25 classes can be naturally divided into one group of 8 classes (group 1) and another of 17 (group 2). If there's reason to believe that:

  1. The binary classification problem of determining whether the signal is a part of group 1 or group 2 can be solved with high accuracy

  2. The classification problem of finding the right class given the group, is significantly easier than the classification problem of finding the right class out of all 25 possible classes

If I were to instead use three classifiers that solve the problem above, is it possible for this architecture to leverage more information than the 1 classifier approach, or is the approach baseless.


1 Answer 1


Yes, this is done all the time. Improved results are often found when splitting features up into different groups, applying a neural net to each different set of features, and then inputting the results from each network into a final network for either classification or function approximation. As an example case, see Figure 2 for which multiple networks are tied together.


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.