August 2019
Intermediate to advanced
342 pages
9h 35m
English
In the following example, adapted from the original code available at https://github.com/eriklindernoren/ML-From-Scratch/blob/master/mlfromscratch/unsupervised_learning/generative_adversarial_network.py (released under the MIT license at https://github.com/eriklindernoren/ML-From-Scratch/blob/master/LICENSE), we see an example of a GAN that is able to artificially generate, from some input noise, the reproduction of handwritten digit images by comparing them against the MNIST dataset (available for download at http://yann.lecun.com/exdb/mnist/).
The activation functions of the GAN's NNs, implemented by the build_generator() and build_discriminator() functions, are both based on Leaky ...