Title: Understanding Long Short Time Memory Neural Networks Post by: kram1032 on August 28, 2015, 02:37:25 AM http://colah.github.io/posts/2015-08-Understanding-LSTMs/ I think with this explanation (and the linked sources) some of you might be able to implement your own neural networks which then could potentially be used for stuff like deep dreaming. (Given that you manage to find appropriate datasets - which are publicly available or you could make your own - and that your computer can actually handle large enough networks in a reasonable time frame) Homebrew Dataset suggestion: Take random escape time pics or (given that these types of networks can actually deal with videos) frame-sets of random polynomials up to order, say, 5, where your polynomial coefficients may also be integers between 0 and 5, and let your network guess which particular iterated polynomial we are currently looking at. Not sure if this would work (in particular, not sure if the zooms into different polynomials actually are different enough to be effectively learned) but it could be a decent start. |