中文
相关论文

相关论文: On the Limits of Latent Reuse in Diffusion Models

200 篇论文

Diffusion models generalize well in practice. However, an optimal diffusion model fully memorizes the training data and therefore fails to generalize, raising the question of what induces generalization in a real diffusion model. We show…

机器学习 · 计算机科学 2026-05-21 Tim Kaiser , Markus Kollmann

Diffusion models learn to restore noisy data, which is corrupted with different levels of noise, by optimizing the weighted sum of the corresponding loss terms, i.e., denoising score matching loss. In this paper, we show that restoring data…

计算机视觉与模式识别 · 计算机科学 2022-04-04 Jooyoung Choi , Jungbeom Lee , Chaehun Shin , Sungwon Kim , Hyunwoo Kim , Sungroh Yoon

Using transfer learning to adapt a pre-trained "source model" to a downstream "target task" can dramatically increase performance with seemingly no downside. In this work, we demonstrate that there can exist a downside after all: bias…

机器学习 · 计算机科学 2022-07-07 Hadi Salman , Saachi Jain , Andrew Ilyas , Logan Engstrom , Eric Wong , Aleksander Madry

Diffusion models learn to denoise data and the trained denoiser is then used to generate new samples from the data distribution. In this paper, we revisit the diffusion sampling process and identify a fundamental cause of sample quality…

机器学习 · 计算机科学 2024-11-05 Yunshu Wu , Yingtao Luo , Xianghao Kong , Evangelos E. Papalexakis , Greg Ver Steeg

High-dimensional data must be highly structured to be learnable. Although the compositional and hierarchical nature of data is often put forward to explain learnability, quantitative measurements establishing these properties are scarce.…

机器学习 · 统计学 2025-03-04 Antonio Sclocchi , Alessandro Favero , Noam Itzhak Levi , Matthieu Wyart

Inverse problems arise in a multitude of applications, where the goal is to recover a clean signal from noisy and possibly (non)linear observations. The difficulty of a reconstruction problem depends on multiple factors, such as the ground…

图像与视频处理 · 电气工程与系统科学 2024-08-21 Zalan Fabian , Berk Tinaz , Mahdi Soltanolkotabi

Latent space model plays a crucial role in network analysis, and accurate estimation of latent variables is essential for downstream tasks such as link prediction. However, the large number of parameters to be estimated presents a…

统计方法学 · 统计学 2025-09-22 Kuangnan Fang , Ruixuan Qin , Xinyan Fan

Diffusion models trained on different, non-overlapping subsets of a dataset often produce strikingly similar outputs when given the same noise seed. We trace this consistency to a simple linear effect: the shared Gaussian statistics across…

机器学习 · 计算机科学 2026-02-04 Binxu Wang , Jacob Zavatone-Veth , Cengiz Pehlevan

Despite their success in image generation, diffusion models can memorize training data, raising serious privacy and copyright concerns. Although prior work has sought to characterize, detect, and mitigate memorization, the fundamental…

计算机视觉与模式识别 · 计算机科学 2025-10-01 Juyeop Kim , Songkuk Kim , Jong-Seok Lee

There is strong empirical evidence that the state-of-the-art diffusion modeling paradigm leads to models that memorize the training set, especially when the training set is small. Prior methods to mitigate the memorization problem often…

机器学习 · 计算机科学 2026-03-03 Kulin Shah , Alkis Kalavasis , Adam R. Klivans , Giannis Daras

When machine learning models are deployed on a test distribution different from the training distribution, they can perform poorly, but overestimate their performance. In this work, we aim to better estimate a model's performance under…

机器学习 · 计算机科学 2020-07-08 Ching-Yao Chuang , Antonio Torralba , Stefanie Jegelka

Real-world datasets are inherently heterogeneous, yet how per-class structural differences and sampling imbalance shape the training dynamics of diffusion models-and potentially exacerbate disparities-remains poorly understood. While models…

When do diffusion models reproduce their training data, and when are they able to generate samples beyond it? A practically relevant theoretical understanding of this interplay between memorization and generalization may significantly…

机器学习 · 计算机科学 2025-08-26 Sam Buchanan , Druv Pai , Yi Ma , Valentin De Bortoli

Diffusion models generate samples by reversing a fixed forward diffusion process. Despite already providing impressive empirical results, these diffusion models algorithms can be further improved by reducing the variance of the training…

机器学习 · 计算机科学 2023-02-20 Yilun Xu , Shangyuan Tong , Tommi Jaakkola

There is a bias in the inference pipeline of most diffusion models. This bias arises from a signal leak whose distribution deviates from the noise distribution, creating a discrepancy between training and inference processes. We demonstrate…

计算机视觉与模式识别 · 计算机科学 2023-10-25 Martin Nicolas Everaert , Athanasios Fitsios , Marco Bocchio , Sami Arpa , Sabine Süsstrunk , Radhakrishna Achanta

Diffusion-based generative models have achieved remarkable success in various domains. It trains a shared model on denoising tasks that encompass different noise levels simultaneously, representing a form of multi-task learning (MTL).…

计算机视觉与模式识别 · 计算机科学 2024-01-02 Hyojun Go , JinYoung Kim , Yunsung Lee , Seunghyun Lee , Shinhyeok Oh , Hyeongdon Moon , Seungtaek Choi

Diffusion models have achieved state-of-the-art synthesis quality on both visual and audio tasks, and recent works further adapt them to textual data by diffusing on the embedding space. In this paper, we conduct systematic studies of the…

计算与语言 · 计算机科学 2024-04-23 Zhujin Gao , Junliang Guo , Xu Tan , Yongxin Zhu , Fang Zhang , Jiang Bian , Linli Xu

The creativity of diffusion models refers to their ability to generate highly realistic images that are different from their training data. Creativity is somewhat surprising since it is known that if the denoiser used in the diffusion model…

计算机视觉与模式识别 · 计算机科学 2026-05-29 Itamar Levine , Yair Weiss

The denoising diffusion model has recently emerged as a powerful generative technique, capable of transforming noise into meaningful data. While theoretical convergence guarantees for diffusion models are well established when the target…

机器学习 · 计算机科学 2025-03-28 Yuchen Liang , Peizhong Ju , Yingbin Liang , Ness Shroff

Designing a single molecule that modulates two targets is a promising strategy for polypharmacology, but it remains substantially harder than standard single-target generation because one candidate must satisfy two binding requirements…

机器学习 · 计算机科学 2026-05-26 Qingyuan Zeng , Pengxiang Cai , Zixin Guan , Ziyang Chen , Anglin Liu , Lang Qin , Xinyao Lai , Jintai Chen