0
$\begingroup$

I'm a medical student, interested on the medical imaging and the neural networks. I have one question:

When I change the training set for a VGG16-CNN algorithm or any other CNN, will the results change?

what I mean is, so many developers developed their own neural networks and then they tested their accuracy

when I use the same neural network 'the same algorithms' but with different training set and with more data, will the accuracy and the results change ?

I'm aiming to use one of the best neural networks to diagnose breast cancer and i'm planning to use a completely different training set, that's why i'm asking this question

$\endgroup$
6
  • 1
    $\begingroup$ Welcome to the site. You may want to start by reading stats.stackexchange.com/tour. This question seems to have two totally different questions in the same post (which is ont ok). The first question seems to be totally off-topic (maybe academia.stackexchange, but I'm afraid it is unclear in the present form and that you need more details about the situation). The second question might be on.topic, but is unclear (what results exactly / why do you expect they would not change)? $\endgroup$ Jun 9, 2018 at 18:15
  • 1
    $\begingroup$ This is still unclear. When you say "the same neural network", do you mean the same architecture and the same parameters, or just the same architecture? If you're training "the same neural network" on a different training set, why would you not expect that "the results change"? What would be the point of training on a new set of data, if not to obtain different (presumably better) results? $\endgroup$ Jun 10, 2018 at 14:52
  • $\begingroup$ Yes I will use a pre-trained neural network "Same architecture and parameters" but i'm aiming to increase the training and test sets [I will train the neural network with more data], will the overall accuracy changes ? $\endgroup$
    – Dalal
    Jun 10, 2018 at 19:03
  • $\begingroup$ If it's still unclear please tell me, I tried my best to read more but still no answer, and I feel like i've only barely scratched the surface ! $\endgroup$
    – Dalal
    Jun 10, 2018 at 19:19
  • $\begingroup$ Yes, the results absolutely depend on the training data you use. $\endgroup$
    – shimao
    Jun 11, 2018 at 15:39

1 Answer 1

1
$\begingroup$

When I change the training set for a VGG16-CNN algorithm or any other CNN, will the results change?

Yes. Machine learning algorithms crucially depend on the data used to train them.

when I use the same neural network 'the same algorithms' but with different training set and with more data, will the accuracy and the results change ?

Yes. Different problems are suited to different solutions. Success of VGG16-CNN on some problem does not imply that it will find success on another one.

$\endgroup$

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.