Figureure 1 · : Overview of the ICR FrameworkFigure 1: Overview of the ICR Framework. Each training image is augmented and passed through the diffusion feature extractor, decomposing representations into an invariant component ?? and a residual ??; their covariances define ICR (a–b). ICR serves as a unified diagnostic: it identifies the optimal noise level for classification tasks, tracks generative quality without sampling, and anticipates memorization onset during training (c).这张图概括 Evaluating the Representation Space of 的整体方法流程。阅读时先看模块之间传递的训练信号,再看作者如何把目标拆成可优化的子问题。Figureure 13 · : Stability of ICR estimates under subsamplingFigure 13: Stability of ICR estimates under subsampling. We evaluate ICR on CIFAR10 using a pretrained EDM model (4095 training samples) and a fixed noise level, varying the number of training samples used to estimate the covariances from $N = 1 6$ up to the full 50K images. As ??increases, the estimated ICR quickly showcase the similar trend close to the full data estimate.这张图来自论文 PDF 的结构化抽取。当前用于辅助理解 Evaluating the Representation Space of 的方法或实验,请结合正文精读段落一起看。