Unsupervised Generative

Unsupervised generative models aim to learn the underlying structure of data without labeled examples, enabling the generation of new data instances similar to the training data. Current research focuses on improving control over the generation process, exploring latent spaces for meaningful manipulation, and developing efficient architectures like GANs and VAEs, including variations with discrete latent variables, for diverse applications. These advancements are impacting fields ranging from anomaly detection in aeronautics to creating synthetic datasets for tasks like tattoo retrieval and improving 3D shape modeling, demonstrating the broad utility of unsupervised generative methods.