{"title":"临界阻尼三阶朗文动力学","authors":"Benjamin Sterling, Monica Bugallo","doi":"arxiv-2409.07697","DOIUrl":null,"url":null,"abstract":"While systems analysis has been studied for decades in the context of control\ntheory, it has only been recently used to improve the convergence of Denoising\nDiffusion Probabilistic Models. This work describes a novel improvement to\nThird- Order Langevin Dynamics (TOLD), a recent diffusion method that performs\nbetter than its predecessors. This improvement, abbreviated TOLD++, is carried\nout by critically damping the TOLD forward transition matrix similarly to\nDockhorn's Critically-Damped Langevin Dynamics (CLD). Specifically, it exploits\neigen-analysis of the forward transition matrix to derive the optimal set of\ndynamics under the original TOLD scheme. TOLD++ is theoretically guaranteed to\nconverge faster than TOLD, and its faster convergence is verified on the Swiss\nRoll toy dataset and CIFAR-10 dataset according to the FID metric.","PeriodicalId":501175,"journal":{"name":"arXiv - EE - Systems and Control","volume":"26 1","pages":""},"PeriodicalIF":0.0000,"publicationDate":"2024-09-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Critically Damped Third-Order Langevin Dynamics\",\"authors\":\"Benjamin Sterling, Monica Bugallo\",\"doi\":\"arxiv-2409.07697\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"While systems analysis has been studied for decades in the context of control\\ntheory, it has only been recently used to improve the convergence of Denoising\\nDiffusion Probabilistic Models. This work describes a novel improvement to\\nThird- Order Langevin Dynamics (TOLD), a recent diffusion method that performs\\nbetter than its predecessors. This improvement, abbreviated TOLD++, is carried\\nout by critically damping the TOLD forward transition matrix similarly to\\nDockhorn's Critically-Damped Langevin Dynamics (CLD). Specifically, it exploits\\neigen-analysis of the forward transition matrix to derive the optimal set of\\ndynamics under the original TOLD scheme. TOLD++ is theoretically guaranteed to\\nconverge faster than TOLD, and its faster convergence is verified on the Swiss\\nRoll toy dataset and CIFAR-10 dataset according to the FID metric.\",\"PeriodicalId\":501175,\"journal\":{\"name\":\"arXiv - EE - Systems and Control\",\"volume\":\"26 1\",\"pages\":\"\"},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2024-09-12\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"arXiv - EE - Systems and Control\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/arxiv-2409.07697\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"arXiv - EE - Systems and Control","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/arxiv-2409.07697","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
While systems analysis has been studied for decades in the context of control
theory, it has only been recently used to improve the convergence of Denoising
Diffusion Probabilistic Models. This work describes a novel improvement to
Third- Order Langevin Dynamics (TOLD), a recent diffusion method that performs
better than its predecessors. This improvement, abbreviated TOLD++, is carried
out by critically damping the TOLD forward transition matrix similarly to
Dockhorn's Critically-Damped Langevin Dynamics (CLD). Specifically, it exploits
eigen-analysis of the forward transition matrix to derive the optimal set of
dynamics under the original TOLD scheme. TOLD++ is theoretically guaranteed to
converge faster than TOLD, and its faster convergence is verified on the Swiss
Roll toy dataset and CIFAR-10 dataset according to the FID metric.