Behind deep-fakes . What are artificial or generative neural networks? Artificial generative adversarial networks , or gans , were introduced by ian goodfellow, et al in his paper “ generative adversarial networks” . These neural networks are an approximation to generative modeling using deep learning methods such as convolutional neural networks . But what is generative modeling all about? Generative modeling of neural networks is an unsupervised learning task , framed within machine learning , which consists of discovering and automatically learning the
regularities or patterns in the input data in such a way that the model can be used to generate or output to new examples that could plausibly have e commerce photo editing service been drawn from the original dataset . Generative model vs discriminator model gans are a clever way to train a generative model , framing the problem as a supervised learning problem with two submodels : the generator model and the discriminator model . Generator model the generator model is in charge of generating the new examples based on the starting data that we have . These machine learning models can learn the statistical latent space of images , music , and even stories .
After this , the model is able to take samples of this space , creating new works of art with characteristics similar to those that the model has seen in its training data ( starting data ) . Discriminator model the discriminator model is in charge of trying to classify the examples created with the generating model as real ( belonging to the problem domain ) or false ( artificially generated ) . After finishing this process , this discriminating model is discarded to remain only with the generator model . How are artificial neural networks trained to perform more accurately? But how do we train these models to be able to