Organiza tus páginas con colecciones
Guarda y categoriza el contenido según tus preferencias.
Las redes generativas adversarias (GAN) son una innovación reciente y emocionante en el aprendizaje automático. Las GAN son modelos generativos: crean instancias de datos nuevas que se parecen a tus datos de entrenamiento. Por ejemplo, las GAN pueden crear imágenes que parecen fotografías de rostros humanos, aunque los rostros no pertenezcan a ninguna persona real. Estas imágenes fueron creadas por un GAN:
Las GAN logran este nivel de realismo combinando un generador, que aprende a producir el resultado objetivo, con un discriminador, que aprende a distinguir los datos reales del resultado del generador. El generador intenta engañar al discriminador, y el discriminador intenta evitar que lo engañen.
En este curso, se abordan los aspectos básicos de las GAN y cómo usar la biblioteca de TF-GAN para crearlas.
[null,null,["Última actualización: 2025-02-26 (UTC)"],[[["\u003cp\u003eGenerative adversarial networks (GANs) are generative models that create new data instances resembling training data, such as images that look like real photographs but are not of actual people.\u003c/p\u003e\n"],["\u003cp\u003eGANs consist of a generator that learns to produce the target output and a discriminator that learns to distinguish real data from generated data, working in tandem to enhance the realism of the output.\u003c/p\u003e\n"],["\u003cp\u003eThis course covers GAN fundamentals, common GAN loss functions, training challenges, and using the TF-GAN library to build GANs, assuming prior knowledge of machine learning and TensorFlow.\u003c/p\u003e\n"],["\u003cp\u003eCompleting Machine Learning Crash Course and having some TensorFlow programming experience are prerequisites for this GANs course.\u003c/p\u003e\n"]]],[],null,["# Introduction\n\n\u003cbr /\u003e\n\nGenerative adversarial networks (GANs) are an exciting recent innovation in\nmachine learning. GANs are *generative* models: they create new data instances\nthat resemble your training data. For example, GANs can create images that look\nlike photographs of human faces, even though the faces don't belong to any real\nperson. These images were created by a GAN:\n\n**Figure 1: Images generated by a\n[GAN created by NVIDIA](https://research.nvidia.com/sites/default/files/pubs/2017-10_Progressive-Growing-of/karras2018iclr-paper.pdf).**\n\nGANs achieve this level of realism by pairing a generator, which learns to\nproduce the target output, with a discriminator, which learns to distinguish\ntrue data from the output of the generator. The generator tries to fool the\ndiscriminator, and the discriminator tries to keep from being fooled.\n\nThis course covers GAN basics, and also how to use the TF-GAN library to create\nGANs.\n| **Course Learning Objectives**\n|\n| - Understand the difference between generative and discriminative models.\n| - Identify problems that GANs can solve.\n| - Understand the roles of the generator and discriminator in a GAN system.\n| - Understand the advantages and disadvantages of common GAN loss functions.\n| - Identify possible solutions to common problems with GAN training.\n| - Use the TF GAN library to make a GAN.\n\nPrerequisites\n-------------\n\nThis course assumes you have:\n\n- Completed [Machine Learning Crash Course](/machine-learning/crash-course) either in-person or self-study.\n- At least a little experience programming with [TensorFlow](https://tensorflow.org)"]]
