Generative Models in Machine Learning and their Applications to Digital Image Generation

Within the Machine Learning field, two types of algorithms can be distinguished if the nature of their outputs as a perspective is acknowledged. Discriminative and generative models associate data with a response and create new data based on a probabilistic distribution of latent variables, respectively. In recent years, significant progress in the Deep Learning subject has been published, which is the study of deep neural networks. Hence, convolutional neural networks have gained significant territory in different tasks regarding image analysis processing. Among the applications of convolutional neural networks, a few of them are Image classification, object detection, instance segmentation, and facial recognition, among others. However, the field of Deep Learning has seen progress not just in these areas but also in the ability of models to generate new images. Thus, a wide variety of generative models for different purposes has been developed; facial images generation of people who do not exist in real life is an example of the latter. Thereupon, this article aims to analyze different generative models for digital image processing and the theoretical aspects that define the generative models in the field of Deep Learning.