Evolving GAN Formulations for Higher Quality Image Synthesis
This work addresses training challenges for researchers and practitioners using GANs for image synthesis, though it is incremental as it builds on existing GAN frameworks.
The paper tackled the problem of GAN training instability and mode collapse by introducing TaylorGAN, a technique that discovers customized loss functions via Taylor expansions and multiobjective evolution, resulting in improved image quality and better performance metrics on an image-to-image translation benchmark.
Generative Adversarial Networks (GANs) have extended deep learning to complex generation and translation tasks across different data modalities. However, GANs are notoriously difficult to train: Mode collapse and other instabilities in the training process often degrade the quality of the generated results, such as images. This paper presents a new technique called TaylorGAN for improving GANs by discovering customized loss functions for each of its two networks. The loss functions are parameterized as Taylor expansions and optimized through multiobjective evolution. On an image-to-image translation benchmark task, this approach qualitatively improves generated image quality and quantitatively improves two independent GAN performance metrics. It therefore forms a promising approach for applying GANs to more challenging tasks in the future.