{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2025,10,10]],"date-time":"2025-10-10T01:05:31Z","timestamp":1760058331479,"version":"build-2065373602"},"reference-count":18,"publisher":"MDPI AG","issue":"4","license":[{"start":{"date-parts":[[2025,3,26]],"date-time":"2025-03-26T00:00:00Z","timestamp":1742947200000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"funder":[{"name":"Innovation in Generative Diffusion Modeling","award":["MTH24000"],"award-info":[{"award-number":["MTH24000"]}]}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":["Entropy"],"abstract":"<jats:p>We investigate diffusion models generating synthetic samples from the probability distribution represented by the ground truth (GT) samples. We focus on how GT sample information is encoded in the score function (SF), computed (not simulated) from the Wiener\u2013Ito linear forward process in the artificial time t\u2208[0\u2192\u221e], and then used as a nonlinear drift in the simulated Wiener\u2013Ito reverse process with t\u2208[\u221e\u21920]. We propose U-Turn diffusion, an augmentation of a pre-trained diffusion model, which shortens the forward and reverse processes to t\u2208[0\u2192Tu] and t\u2208[Tu\u21920]. The U-Turn reverse process is initialized at Tu with a sample from the probability distribution of the forward process (initialized at t=0 with a GT sample) ensuring a detailed balance relation between the shortened forward and reverse processes. Our experiments on the class-conditioned SF of the ImageNet dataset and the multi-class, single SF of the CIFAR-10 dataset reveal a critical Memorization Time Tm, beyond which generated samples diverge from the GT sample used to initialize the U-Turn scheme, and a Speciation Time Ts, where for Tu&gt;Ts&gt;Tm, samples begin representing different classes. We further examine the role of SF nonlinearity through a Gaussian Test, comparing empirical and Gaussian-approximated U-Turn auto-correlation functions and showing that the SF becomes effectively affine for t&gt;Ts and approximately affine for t\u2208[Tm,Ts].<\/jats:p>","DOI":"10.3390\/e27040343","type":"journal-article","created":{"date-parts":[[2025,3,26]],"date-time":"2025-03-26T09:13:26Z","timestamp":1742980406000},"page":"343","update-policy":"https:\/\/doi.org\/10.3390\/mdpi_crossmark_policy","source":"Crossref","is-referenced-by-count":0,"title":["U-Turn Diffusion"],"prefix":"10.3390","volume":"27","author":[{"given":"Hamidreza","family":"Behjoo","sequence":"first","affiliation":[{"name":"Program in Applied Mathematics, Department of Mathematics, University of Arizona, Tucson, AZ 85721, USA"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-6758-515X","authenticated-orcid":false,"given":"Michael","family":"Chertkov","sequence":"additional","affiliation":[{"name":"Program in Applied Mathematics, Department of Mathematics, University of Arizona, Tucson, AZ 85721, USA"}]}],"member":"1968","published-online":{"date-parts":[[2025,3,26]]},"reference":[{"key":"ref_1","unstructured":"Song, Y., Sohl-Dickstein, J., Kingma, D.P., Kumar, A., Ermon, S., and Poole, B. (2021, January 4). Score-Based Generative Modeling through Stochastic Differential Equations. Proceedings of the International Conference on Learning Representations, Vienna, Austria."},{"key":"ref_2","doi-asserted-by":"crossref","unstructured":"Rombach, R., Blattmann, A., Lorenz, D., Esser, P., and Ommer, B. (2022, January 18\u201324). High-resolution image synthesis with latent diffusion models. Proceedings of the IEEE\/CVF Conference on Computer Vision and Pattern Recognition (CVPR), New Orleans, LA, USA.","DOI":"10.1109\/CVPR52688.2022.01042"},{"key":"ref_3","unstructured":"Ho, J., Jain, A., and Abbeel, P. (2020, January 6\u201312). Denoising Diffusion Probabilistic Models. Proceedings of the Advances in Neural Information Processing Systems, Vancouver, BC, Canada."},{"key":"ref_4","doi-asserted-by":"crossref","first-page":"313","DOI":"10.1016\/0304-4149(82)90051-5","article-title":"Reverse-time diffusion equation models","volume":"12","author":"Anderson","year":"1982","journal-title":"Stoch. Process. Their Appl."},{"key":"ref_5","unstructured":"Karras, T., Aittala, M., Aila, T., and Laine, S. (December, January 28). Elucidating the Design Space of Diffusion-Based Generative Models. Proceedings of the Advances in Neural Information Processing Systems, New Orleans, LA, USA."},{"key":"ref_6","doi-asserted-by":"crossref","first-page":"9957","DOI":"10.1038\/s41467-024-54281-3","article-title":"Dynamical regimes of diffusion models","volume":"15","author":"Biroli","year":"2024","journal-title":"Nat. Commun."},{"key":"ref_7","doi-asserted-by":"crossref","unstructured":"Raya, G., and Ambrogioni, L. (2023, January 10\u201316). Spontaneous symmetry breaking in generative diffusion models. Proceedings of the Thirty-Seventh Conference on Neural Information Processing Systems, Orleans, LA, USA.","DOI":"10.1088\/1742-5468\/ad64bd"},{"key":"ref_8","doi-asserted-by":"crossref","first-page":"093402","DOI":"10.1088\/1742-5468\/acf8ba","article-title":"Generative diffusion in very large dimensions","volume":"2023","author":"Biroli","year":"2023","journal-title":"J. Stat. Mech. Theory Exp."},{"key":"ref_9","doi-asserted-by":"crossref","unstructured":"Sclocchi, A., Favero, A., and Wyart, M. (2024). A phase transition in diffusion models reveals the hierarchical nature of data. arXiv.","DOI":"10.1073\/pnas.2408799121"},{"key":"ref_10","unstructured":"Luzi, L., Mayer, P.M., Casco-Rodriguez, J., Siahkoohi, A., and Baraniuk, R. (2024). Boomerang: Local sampling on image manifolds using diffusion models. Trans. Mach. Learn. Res."},{"key":"ref_11","unstructured":"Sohl-Dickstein, J., Weiss, E., Maheswaranathan, N., and Ganguli, S. (2015, January 7\u20139). Deep Unsupervised Learning using Nonequilibrium Thermodynamics. Proceedings of the 32nd International Conference on Machine Learning, Lille, France."},{"key":"ref_12","unstructured":"Song, Y., and Ermon, S. (2019, January 8\u201314). Generative Modeling by Estimating Gradients of the Data Distribution. Proceedings of the Advances in Neural Information Processing Systems, Vancouver, BC, Canada."},{"key":"ref_13","unstructured":"Wang, B., and Vastola, J. (2024). The unreasonable effectiveness of Gaussian score approximation for diffusion models and its applications. Trans. Mach. Learn. Res."},{"key":"ref_14","first-page":"6626","article-title":"GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium","volume":"30","author":"Heusel","year":"2017","journal-title":"Adv. Neural Inf. Process. Syst. (NeurIPS)"},{"key":"ref_15","doi-asserted-by":"crossref","first-page":"68","DOI":"10.1080\/01621459.1951.10500769","article-title":"The Kolmogorov-Smirnov test for goodness of fit","volume":"46","author":"Massey","year":"1951","journal-title":"J. Am. Stat. Assoc."},{"key":"ref_16","doi-asserted-by":"crossref","first-page":"279","DOI":"10.1214\/aoms\/1177730256","article-title":"Table for Estimating the Goodness of Fit of Empirical Distributions","volume":"19","author":"Smirnov","year":"1948","journal-title":"Ann. Math. Stat."},{"key":"ref_17","first-page":"83","article-title":"Sulla determinazione empirica di una legge di distribuzione","volume":"4","author":"Kolmogorov","year":"1933","journal-title":"Giornale Dell\u2019Istituto Ital. Degli Attuari"},{"key":"ref_18","unstructured":"Pierret, E., and Galerne, B. (2024). Diffusion models for Gaussian distributions: Exact solutions and Wasserstein errors. arXiv."}],"container-title":["Entropy"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/www.mdpi.com\/1099-4300\/27\/4\/343\/pdf","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2025,10,9]],"date-time":"2025-10-09T17:02:17Z","timestamp":1760029337000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.mdpi.com\/1099-4300\/27\/4\/343"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2025,3,26]]},"references-count":18,"journal-issue":{"issue":"4","published-online":{"date-parts":[[2025,4]]}},"alternative-id":["e27040343"],"URL":"https:\/\/doi.org\/10.3390\/e27040343","relation":{},"ISSN":["1099-4300"],"issn-type":[{"type":"electronic","value":"1099-4300"}],"subject":[],"published":{"date-parts":[[2025,3,26]]}}}