1
$\begingroup$

I'm learning neural networks, and I'd like to give myself simple exercises and try to train neural networks to solve them.

I know that classifying hand-written digits is a classic problem, but I'm hoping for something much simpler than that. No image recognition, no thousand of input neurons.

One exercise I thought of, is to get a number and find its modulo when dividing by 7. But it was explained to me that this is actually a difficult task for a neural network beginner for various reasons.

Does anyone have ideas for a very simple exercise/training idea? Or a link to an existing list of exercises?

$\endgroup$

4 Answers 4

0
$\begingroup$

I suggest you to practice with small datasets such as:

These are good for classification tasks. For regression tasks, I suggest you something like the Boston housing dataset, downloadable directly from sklearn.

They are all very small datasets that can be trained in few minutes (if not seconds). They are really useful to get used to model tuning, minibatch training, and the syntax of course.

In my very first experiment with Neural Networks I trained one on the breast cancer dataset, a simple dense network that trained in few seconds on an average laptop. You can take a look here.

$\endgroup$
1
  • $\begingroup$ This is great, thank you! $\endgroup$
    – Ram Rachum
    Commented Jun 4, 2020 at 17:39
1
$\begingroup$

I'm learning myself too and I found kaggle.com and excellent resource for learning and practice, they have lots of data sets available for you to try your own models.

There is also a learn section with lots of topics for beginners

Hope that helps.

$\endgroup$
2
  • $\begingroup$ Thanks Ricardo. I see that's an entire platform / course, can you tell me the actual problem that could be relevant? i.e. what is the dataset and desired labels to classify to? $\endgroup$
    – Ram Rachum
    Commented Jun 2, 2020 at 18:34
  • $\begingroup$ There is no particular case for that I just point it out as a learning resource, regardless of the data size I found learning is segmented in to different categories, learn how to prep data, learn how to build the network and learn how to process the results. $\endgroup$ Commented Jun 2, 2020 at 19:01
0
$\begingroup$

Here's one exercise I came up with:

Given a set of 6 English letters, predict whether there is a 6-letter English word that is composed of these letters.

I solved it to 90% accuracy without knowing much about neural networks.

I'll be happy to get more simple exercise problems!

$\endgroup$
0
$\begingroup$

I am not sure if you have heard about it but mimicking boolean functions using neural networks is one of the most basic problems. Here is a link if you want to know more or here.

$\endgroup$

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.