Even as an outsider to neural network development, they still seem to be a hot topic... I get lots of projects I see being starred on my GitHub homepage relating to neural networks.

These projects look interesting, they come with lots of visuals. Typically with images being mixed together in various ways. Their patterns and colors being interchanged... Unfortunately though, this is the only thing I see happening with neural networks in these projects, and I don't really understand why so much research and effort would go into that one topic, to make weird images... So there must be more uses that I'm not aware of.

What are the practical uses of a neural network?

  • Paging @jimmyhoffa... – user22815 Apr 26 '16 at 3:41
  • I've written a generic game player that learns how to play arbitrary games using neural networks (which is based off of a similar game playing neural-network, TD-Gammon) – Nathan Merrill Apr 26 '16 at 4:12
  • 3
    Are you looking for something else than what you'd find by checking out an obvious source? – COME FROM Apr 26 '16 at 6:15
  • Neural Networks are essentially pattern recognition engines. They learn to map a set of inputs to a set of outputs when the set of inputs is not complete. – Steven Burnap Apr 26 '16 at 23:56

The practical uses of a neural network is pretty much everything.

  1. Recognition / detection in vision
  2. Artificial intelligence in games
  3. Classification
  4. ...

In short : neural network can do pretty much everything as long you're able to get enough data and some efficient machine to get the right parameters. My professor told me once that some competition between schools in neural network turned in the end to the one having the most efficient computers. Because neural network for real problems need a lot of calculation power for the learning phase.

However neural network have some drawbacks, when it come to classification : the value of the parameters of the networks means pretty much nothing to a human.

There is other algorithm that produces results that you can read. I remember of one but only the French name, it's a Lasso regression with an absolute value of a lambda variable. Though it won't ever beat a neural network, you can read the results from the learning and find unnecessary data in order to exclude them. So it can in the end improve the results of your neural network or whatever you use after (SVM, kernels ...).

While neural networks are really powerful, they're not magic and won't beat every other algorithm in all fields. For example, the capacity of learning really fast might make a Support vector machine a better fit, since some can learn and process classification with incredible speed and accuracy.


I think the biggest successes have been in classification problems like speech recognition and image classification (facial recognition etc)

  • This is way too short and incomplete to be a real answer. – Steven Burnap Apr 26 '16 at 23:54

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy