From GANs to VAEs: A Survey of Generative Models in Artificial Intelligence

Abstract

Computational methods that can produce seemingly original, meaningful content—like writing, images, or audio—from training data are referred to as “generative AI.” the broad adoption of this technology, as demonstrated by GPT-4 and Dall-E 2. A type of artificial intelligence known as “generative AI” can create new text, images, audio, and video content on its own. Filling in the gaps in the metaverse's evolution, generative AI offers creative methods for content creation in the metaverse. GANs work through a competitive game between a generator and a discriminator, whereas VAEs use probabilistic latent variables and an encoder-decoder architecture. This survey examines GANs and VAEs in depth, including their designs, training techniques, applications, and recent breakthroughs. Furthermore, it dives into each model's distinct strengths and shortcomings, providing insights into when and how to use them effectively. Additionally, it addresses hybrid approaches that combine the characteristics of GANs and VAEs, as well as emerging trends and problems in the field of generative modeling. Products like Chat GPT have the power to improve search results, change how information is generated and presented, and open new channels for internet traffic. This is anticipated to have a substantial impact on conventional search engine offerings, speeding up industry innovation and modernization. To increase performance and safety, artificial intelligence (AI) is utilized to mimic and create situations for testing and training.