In this section, we explore methods for scaling up GAN training to reap the performance benefits of larger models and larger batches. As a baseline, we employ the SA-GAN architecture of Zhang et al. (2018), which uses the hinge loss (Lim & …where $\mathbf{1}$ denotes a matrix with all elements set to 1. We sweep $\beta$ values and select $10^{-4}$, finding this small added penalty sufficient to improve the likelihood that our models will be amenable to truncation. Across runs …
Karras et al.
other · 2 mentions across 1 reading
In this course
Karras et al. are foundational researchers in scaling generative adversarial networks, referenced here for their work on GAN architectures and training techniques that enable high-quality image synthesis. The course readings invoke their methods as baselines for exploring how to push GAN performance through larger models and batch sizes, establishing the technical scaffolding upon which subsequent innovations in generative modeling are built. Their work anchors the seminar's discussion of how computational scale intersects with the aesthetic and technical possibilities of AI-generated imagery.
Mentioned in 1 reading
Appears alongside
People mentioned in the same passages — sorted by co-occurrence weight.