I keep reading that convolution neural net (CNN) performs best with lots and lots (100k+) of data. Is there any rule of thumb, or lower limit for data size during the grid search phase?

For example, if I run a CNN with 100 data points, vary just one parameter (say add an extra layer, or increase a filter size), and get better results, can I reasonably expect better results with those parameters during the actual training phase?

  • $\begingroup$ It's wrong that you need a lot of images to train a conv-net... I had only trained them with 22 images as trainset and 7 as validation and test.... that also works $\endgroup$ – Aditya Apr 5 '18 at 2:35

If you use pre-trained weights, you need significantly lesser data as the initial layers have already learned from a ton of data and you just need to fine tune the later ones.

What you said is not true, you can train on CIFAR10 and get 90%+ and that is not 100k+. It depends on the complexity of the data and how similar the features are. If they are easily Seperable -less data. If the disctintions are harder then the model needs a lot of examples to figure out which of features are seperate.

I would say you could IF you sample was representive of the population.

| improve this answer | |

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.