A javascript implementation of one of Michael Nielsen's neural network algorithms http://neuralnetworksanddeeplearning.com/.
The original purpose of this code was to experiment with classifying MNIST digits in the browser in a reasonable amount of time and with low memory. I ended up deciding it was not worth the value, people would not want to wait even 30 seconds per run let alone up to an hour.
Running a network with 30 hidden nodes took this program almost 55 minutes, while Michael Nielsen's algorithm took less then a minute using Python and Numpy. Numpy handles matrices in a high performant way using a C (or C++) implementation. This JS library includes some element-wise matrix functions which may be the bottleneck in the algorithm, although more testing would be necessary.
This code served as a nice learning experience, but should not be used by anyone.