세균맨

Here I introduce the generalization error bound of the Domain generalization problem, which is the test domain—or style, sometimes—differs from the training domain.PreliminariesNotations$X \in \mathcal{X} \subset \mathbb{R}^d, Y\in \mathcal{Y} \subset \mathbb{R}$ : Common input and target space$P^i_{XY}$: Data distribution of the i'th domain$S^i\sim P^i_{XY}$: Samples for the i'th domain$\epsilo..

ICLR2023 Notable top 25% (10/8/8/5), Stable diffusion 3Motivation어떤 data-distribution 에서 simple-distribution (e.g. standard gaussian) 으로 변화하는 path (e.g. forward-diffusion process) 를 좀 더 잘 정의해서, 그것의 inverse (image generation via the diffusion model) 또한 더 잘 되도록 하고싶다 Simple diffusion process (adding a “simple“ gaussian noise) leads to rather confined space of sampling probability paths. (라고 표현하고 있는..