What are the standard (or baseline) problems (or at least common ones) for CNNs and LSTMs? As an example, for a feed-forward neural net, a common problem is the XOR problem.

Is there a standard problem like this for CNNs and LSTMs? I think for a CNN the standard test is to try it on MNIST, but I'm not sure of an LSTM.

  • $\begingroup$ for CNN its mnist $\endgroup$
    – Eka
    Commented Oct 14, 2019 at 5:20

2 Answers 2


It's more domain- or task-specific. There is no obvious baseline anymore because these models and this field has evolved into too large of an ecosystem. Nonetheless, I'll list a couple of notorious examples below.

Image classification:

  • ImageNet


  • PascalVOC
  • COCO
  • CityScapes

Pose estimation:

  • MPII

Text classification:

  • IMDB
  • yelp

Question answering:

  • SQuAD


  • WMT

This is just a taste. There are tons more both in each category and the number of categories, a good source is the Papers with Code website.

Therefore, there is no single standard problem, given that there are too many that all in one shape or form use CNNs or RNNs (or others).


I think for LSTM, I have not come across a standard test, but when I started, I tried like this.

Generate a sequence of numbers like, [0,1,2,3,4],[1,2,3,4,5]..... as the dataset and then your labels would be [5,6,.....].Train this using an LSTM network.This would be a good way to understand the parameters involved and by changing the number of layers, and different parameters you can easily check how it works.

of course, mnist is the test for CNNs

  • $\begingroup$ Yeh I did something similar (except using text, so I had a string of "the quick brown fox..." and used each character as a 1 hot vector) and I know it works fine, but the big problem is when testing your model against others (if you try to make an improvement somehow), it's good to use a standard so when trying to argue that x is better, you have something to compare that everyone is familiar with. $\endgroup$
    – Recessive
    Commented Oct 14, 2019 at 5:38

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .