{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,3,22]],"date-time":"2026-03-22T06:02:40Z","timestamp":1774159360697,"version":"3.50.1"},"reference-count":42,"publisher":"MDPI AG","issue":"3","license":[{"start":{"date-parts":[[2026,3,20]],"date-time":"2026-03-20T00:00:00Z","timestamp":1773964800000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"funder":[{"name":"Wallenberg AI, Autonomous Systems and Software Program"},{"DOI":"10.13039\/501100004063","name":"Knut and Alice Wallenberg Foundation","doi-asserted-by":"crossref","id":[{"id":"10.13039\/501100004063","id-type":"DOI","asserted-by":"crossref"}]},{"name":"National Academic Infrastructure for Supercomputing in Sweden"},{"DOI":"10.13039\/501100004359","name":"Swedish Research Council","doi-asserted-by":"publisher","award":["2022-06725"],"award-info":[{"award-number":["2022-06725"]}],"id":[{"id":"10.13039\/501100004359","id-type":"DOI","asserted-by":"publisher"}]}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":["Entropy"],"abstract":"<jats:p>Diffusion models can be parameterized in terms of either score or energy function. The energy parameterization is attractive as it enables sampling procedures such as Markov Chain Monte Carlo (MCMC) that incorporates a Metropolis\u2013Hastings (MH) correction step based on energy differences between proposed samples. Such corrections can significantly improve sampling quality, particularly in the context of model composition, where pre-trained models are combined to generate samples from novel distributions. Score-based diffusion models, on the other hand, are more widely adopted and come with a rich ecosystem of pre-trained models. However, they do not, in general, define an underlying energy function, making MH-based sampling inapplicable. In this work, we address this limitation by retaining score parameterization and introducing a novel MH-like acceptance rule based on line integration of the score function. This allows the reuse of existing diffusion models while still combining the reverse process with various MCMC techniques, viewed as an instance of annealed MCMC. Through experiments on synthetic and real-world data, we show that our MH-like samplers yield relative improvements of similar magnitude to those observed with energy-based models, without requiring explicit energy parameterization.<\/jats:p>","DOI":"10.3390\/e28030351","type":"journal-article","created":{"date-parts":[[2026,3,20]],"date-time":"2026-03-20T11:44:46Z","timestamp":1774007086000},"page":"351","update-policy":"https:\/\/doi.org\/10.3390\/mdpi_crossmark_policy","source":"Crossref","is-referenced-by-count":0,"title":["MCMC Correction of Score-Based Diffusion Models for Model Composition"],"prefix":"10.3390","volume":"28","author":[{"ORCID":"https:\/\/orcid.org\/0009-0001-1185-5018","authenticated-orcid":false,"given":"Anders","family":"Sj\u00f6berg","sequence":"first","affiliation":[{"name":"Fraunhofer-Chalmers Centre, SE-412 88 Gothenburg, Sweden"},{"name":"Department of Electrical Engineering, Chalmers University of Technology, SE-412 96 Gothenburg, Sweden"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-2790-8775","authenticated-orcid":false,"given":"Jakob","family":"Lindqvist","sequence":"additional","affiliation":[{"name":"Department of Electrical Engineering, Chalmers University of Technology, SE-412 96 Gothenburg, Sweden"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Magnus","family":"\u00d6nnheim","sequence":"additional","affiliation":[{"name":"Fraunhofer-Chalmers Centre, SE-412 88 Gothenburg, Sweden"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-6612-8037","authenticated-orcid":false,"given":"Mats","family":"Jirstrand","sequence":"additional","affiliation":[{"name":"Fraunhofer-Chalmers Centre, SE-412 88 Gothenburg, Sweden"},{"name":"Department of Electrical Engineering, Chalmers University of Technology, SE-412 96 Gothenburg, Sweden"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-0206-9186","authenticated-orcid":false,"given":"Lennart","family":"Svensson","sequence":"additional","affiliation":[{"name":"Department of Electrical Engineering, Chalmers University of Technology, SE-412 96 Gothenburg, Sweden"}],"role":[{"role":"author","vocabulary":"crossref"}]}],"member":"1968","published-online":{"date-parts":[[2026,3,20]]},"reference":[{"key":"ref_1","unstructured":"Brock, A., Donahue, J., and Simonyan, K. (2019, January 6\u20139). Large Scale GAN Training for High Fidelity Natural Image Synthesis. Proceedings of the International Conference on Learning Representations (ICLR), New Orleans, LA, USA."},{"key":"ref_2","first-page":"1877","article-title":"Language models are few-shot learners","volume":"33","author":"Brown","year":"2020","journal-title":"Adv. Neural Inf. Process. Syst."},{"key":"ref_3","first-page":"6840","article-title":"Denoising diffusion probabilistic models","volume":"33","author":"Ho","year":"2020","journal-title":"Adv. Neural Inf. Process. Syst."},{"key":"ref_4","doi-asserted-by":"crossref","first-page":"1092","DOI":"10.1126\/science.abq1158","article-title":"Competition-level code generation with alphacode","volume":"378","author":"Li","year":"2022","journal-title":"Science"},{"key":"ref_5","first-page":"36479","article-title":"Photorealistic text-to-image diffusion models with deep language understanding","volume":"35","author":"Saharia","year":"2022","journal-title":"Adv. Neural Inf. Process. Syst."},{"key":"ref_6","doi-asserted-by":"crossref","first-page":"102872","DOI":"10.1016\/j.media.2023.102872","article-title":"Adaptive diffusion priors for accelerated MRI reconstruction","volume":"88","author":"Dar","year":"2023","journal-title":"Med. Image Anal."},{"key":"ref_7","doi-asserted-by":"crossref","unstructured":"Wynn, J., and Turmukhambetov, D. (2023, January 18\u201322). DiffusioNeRF: Regularizing neural radiance fields with denoising diffusion models. Proceedings of the IEEE\/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Vancouver, BC, Canada.","DOI":"10.1109\/CVPR52729.2023.00407"},{"key":"ref_8","unstructured":"Sohl-Dickstein, J., Weiss, E., Maheswaranathan, N., and Ganguli, S. (2015, January 6\u201311). Deep unsupervised learning using nonequilibrium thermodynamics. Proceedings of the International Conference on Machine Learning (ICML), Lille, France."},{"key":"ref_9","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 32, Vancouver, BC, Canada."},{"key":"ref_10","first-page":"8780","article-title":"Diffusion models beat GANs on image synthesis","volume":"34","author":"Dhariwal","year":"2021","journal-title":"Adv. Neural Inf. Process. Syst."},{"key":"ref_11","doi-asserted-by":"crossref","unstructured":"L\u00fcdke, D., Bilo\u0161, M., Shchur, O., Lienen, M., and G\u00fcnnemann, S. (2023, January 10\u201316). Add and Thin: Diffusion for Temporal Point Processes. Proceedings of the Thirty-Seventh Conference on Neural Information Processing Systems (NeurIPS), New Orleans, LA, USA.","DOI":"10.52202\/075280-2480"},{"key":"ref_12","unstructured":"Wang, K., Xu, Z., Zhou, Y., Zang, Z., Darrell, T., Liu, Z., and You, Y. (2024). Neural Network Diffusion. arXiv."},{"key":"ref_13","doi-asserted-by":"crossref","first-page":"79","DOI":"10.1162\/neco.1991.3.1.79","article-title":"Adaptive mixtures of local experts","volume":"3","author":"Jacobs","year":"1991","journal-title":"Neural Comput."},{"key":"ref_14","doi-asserted-by":"crossref","first-page":"1771","DOI":"10.1162\/089976602760128018","article-title":"Training products of experts by minimizing contrastive divergence","volume":"14","author":"Hinton","year":"2002","journal-title":"Neural Comput."},{"key":"ref_15","unstructured":"Mayraz, G., and Hinton, G.E. (December, January 27). Recognizing hand-written digits using hierarchical products of experts. Proceedings of the 13th International Conference on Neural Information Processing Systems (NeurIPS), Denver, CO, USA."},{"key":"ref_16","doi-asserted-by":"crossref","unstructured":"Liu, N., Li, S., Du, Y., Torralba, A., and Tenenbaum, J.B. (2022). Compositional visual generation with composable diffusion models. Proceedings of the Computer Vision\u2013ECCV 2022: 17th European Conference, Tel Aviv, Israel, 23\u201327 October 2022, Springer. Proceedings, Part XXVII.","DOI":"10.1007\/978-3-031-19790-1_26"},{"key":"ref_17","unstructured":"Ho, J., and Salimans, T. (2021, January 14). Classifier-Free Diffusion Guidance. Proceedings of the NeurIPS 2021 Workshop on Deep Generative Models and Downstream Applications, Virtual Event."},{"key":"ref_18","first-page":"8489","article-title":"Reduce, Reuse, Recycle: Compositional Generation with Energy-Based Diffusion Models and MCMC","volume":"Volume 202","author":"Du","year":"2023","journal-title":"Proceedings of the 40th International Conference on Machine Learning (ICML), Honolulu, HI, USA, 23\u201329 July 2023"},{"key":"ref_19","unstructured":"Aghajanyan, A., Yu, L., Conneau, A., Hsu, W.N., Hambardzumyan, K., Zhang, S., Roller, S., Goyal, N., Levy, O., and Zettlemoyer, L. (2023). Scaling laws for generative mixed-modal language models. Proceedings of the 40th International Conference on Machine Learning (ICML), Honolulu, HI, USA, 23\u201329 July 2023, Proceedings of Machine Learning Research."},{"key":"ref_20","unstructured":"Song, Y., Sohl-Dickstein, J., Kingma, D.P., Kumar, A., Ermon, S., and Poole, B. (2021, January 3\u20137). Score-Based Generative Modeling through Stochastic Differential Equations. Proceedings of the International Conference on Learning Representations (ICLR), Virtual Event."},{"key":"ref_21","doi-asserted-by":"crossref","first-page":"337","DOI":"10.1023\/A:1023562417138","article-title":"Langevin Diffusions and Metropolis\u2013Hastings Algorithms","volume":"4","author":"Roberts","year":"2002","journal-title":"Methodol. Comput. Appl. Probab."},{"key":"ref_22","doi-asserted-by":"crossref","first-page":"216","DOI":"10.1016\/0370-2693(87)91197-X","article-title":"Hybrid Monte Carlo","volume":"195","author":"Duane","year":"1987","journal-title":"Phys. Lett. B"},{"key":"ref_23","doi-asserted-by":"crossref","first-page":"1087","DOI":"10.1063\/1.1699114","article-title":"Equation of State Calculations by Fast Computing Machines","volume":"21","author":"Metropolis","year":"1953","journal-title":"J. Chem. Phys."},{"key":"ref_24","doi-asserted-by":"crossref","first-page":"97","DOI":"10.1093\/biomet\/57.1.97","article-title":"Monte Carlo Sampling Methods Using Markov Chains and Their Applications","volume":"57","author":"Hastings","year":"1970","journal-title":"Biometrika"},{"key":"ref_25","unstructured":"Salimans, T., and Ho, J. (2021, January 7). Should EBMs model the energy or the score?. Proceedings of the Energy-Based Models Workshop at ICLR 2021, Virtual Event."},{"key":"ref_26","doi-asserted-by":"crossref","unstructured":"LeCun, Y., Chopra, S., Hadsell, R., Ranzato, M., and Huang, F. (2006). A tutorial on energy-based learning. Predicting Structured Data, MIT Press.","DOI":"10.7551\/mitpress\/7443.003.0014"},{"key":"ref_27","doi-asserted-by":"crossref","first-page":"341","DOI":"10.2307\/3318418","article-title":"Exponential convergence of Langevin distributions and their discrete approximations","volume":"2","author":"Roberts","year":"1996","journal-title":"Bernoulli"},{"key":"ref_28","first-page":"639","article-title":"MCMC variational inference via uncorrected Hamiltonian annealing","volume":"34","author":"Geffner","year":"2021","journal-title":"Adv. Neural Inf. Process. Syst."},{"key":"ref_29","first-page":"591","article-title":"Comments on \u201cRepresentations of knowledge in complex systems\u201d by U. Grenander and M. I. Miller","volume":"56","author":"Besag","year":"1994","journal-title":"J. R. Stat. Soc. Ser. B (Methodol.)"},{"key":"ref_30","doi-asserted-by":"crossref","unstructured":"Neal, R.M., Diggle, P., and Fienberg, S. (1996). Bayesian Learning for Neural Networks, Springer.","DOI":"10.1007\/978-1-4612-0745-0"},{"key":"ref_31","unstructured":"Chung, H., Kim, J., Mccann, M.T., Klasky, M.L., and Ye, J.C. (2023, January 1\u20135). Diffusion Posterior Sampling for General Noisy Inverse Problems. Proceedings of the The Eleventh International Conference on Learning Representations (ICLR), Kigali, Rwanda."},{"key":"ref_32","first-page":"8633","article-title":"Video Diffusion Models","volume":"Volume 35","author":"Koyejo","year":"2022","journal-title":"Proceedings of the Advances in Neural Information Processing Systems (NeurIPS)"},{"key":"ref_33","doi-asserted-by":"crossref","first-page":"125","DOI":"10.1023\/A:1008923215028","article-title":"Annealed importance sampling","volume":"11","author":"Neal","year":"2001","journal-title":"Stat. Comput."},{"key":"ref_34","doi-asserted-by":"crossref","first-page":"141","DOI":"10.1109\/MSP.2012.2211477","article-title":"The MNIST Database of Handwritten Digit Images for Machine Learning Research [Best of the Web]","volume":"29","author":"Deng","year":"2012","journal-title":"IEEE Signal Process. Mag."},{"key":"ref_35","unstructured":"Bradbury, J., Frostig, R., Hawkins, P., Johnson, M.J., Leary, C., Maclaurin, D., Necula, G., Paszke, A., VanderPlas, J., and Wanderman-Milne, S. (2025, January 04). JAX: Composable Transformations of Python+NumPy Programs. Version 0.4.30. Available online: https:\/\/github.com\/google\/jax."},{"key":"ref_36","unstructured":"Krizhevsky, A., and Hinton, G. (2009). Learning Multiple Layers of Features from Tiny Images, University of Toronto. Technical Report."},{"key":"ref_37","doi-asserted-by":"crossref","unstructured":"Deng, J., Dong, W., Socher, R., Li, L.J., Li, K., and Fei-Fei, L. (2009, January 20\u201325). Imagenet: A large-scale hierarchical image database. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Miami, FL, USA.","DOI":"10.1109\/CVPR.2009.5206848"},{"key":"ref_38","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."},{"key":"ref_39","unstructured":"Simonyan, K., and Zisserman, A. (2014). Very deep convolutional networks for large-scale image recognition. arXiv."},{"key":"ref_40","doi-asserted-by":"crossref","unstructured":"Radosavovic, I., Kosaraju, R.P., Girshick, R., He, K., and Doll\u00e1r, P. (2020, January 13\u201319). Designing network design spaces. Proceedings of the IEEE\/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Seattle, WA, USA.","DOI":"10.1109\/CVPR42600.2020.01044"},{"key":"ref_41","unstructured":"Horvat, C., and Pfister, J.P. (2024). On gauge freedom, conservativity and intrinsic dimensionality estimation in diffusion models. arXiv."},{"key":"ref_42","first-page":"8162","article-title":"Improved denoising diffusion probabilistic models","volume":"Volume 139","author":"Nichol","year":"2021","journal-title":"Proceedings of the International Conference on Machine Learning (ICML), Virtual Event, 18\u201324 July 2021"}],"container-title":["Entropy"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/www.mdpi.com\/1099-4300\/28\/3\/351\/pdf","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,3,22]],"date-time":"2026-03-22T05:28:41Z","timestamp":1774157321000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.mdpi.com\/1099-4300\/28\/3\/351"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,3,20]]},"references-count":42,"journal-issue":{"issue":"3","published-online":{"date-parts":[[2026,3]]}},"alternative-id":["e28030351"],"URL":"https:\/\/doi.org\/10.3390\/e28030351","relation":{},"ISSN":["1099-4300"],"issn-type":[{"value":"1099-4300","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,3,20]]}}}