Abstract
Successive image generation using cyclic transformations is demonstrated byextending the CycleGAN model to transform images among three differentcategories. Repeated application of the trained generators produces sequencesof images that transition among the different categories. The generated imagesequences occupy a more limited region of the image space compared with theoriginal training dataset. Quantitative evaluation using precision and recallmetrics indicates that the generated images have high quality but reduceddiversity relative to the training dataset. Such successive generationprocesses are characterized as chaotic dynamics in terms of dynamical systemtheory. Positive Lyapunov exponents estimated from the generated trajectoriesconfirm the presence of chaotic dynamics, with the Lyapunov dimension of theattractor found to be comparable to the intrinsic dimension of the trainingdata manifold. The results suggest that chaotic dynamics in the image spacedefined by the deep generative model contribute to the diversity of thegenerated images, constituting a novel approach for multi-class imagegeneration. This model can be interpreted as an extension of classicalassociative memory to perform hetero-association among image categories.