{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,4,23]],"date-time":"2026-04-23T05:18:08Z","timestamp":1776921488496,"version":"3.51.2"},"reference-count":42,"publisher":"Frontiers Media SA","license":[{"start":{"date-parts":[[2024,8,26]],"date-time":"2024-08-26T00:00:00Z","timestamp":1724630400000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"content-domain":{"domain":["frontiersin.org"],"crossmark-restriction":true},"short-container-title":["Front. Comput. Sci."],"abstract":"<jats:p>Many deep generative models, such as variational autoencoders (VAEs) and generative adversarial networks (GANs), learn an immersion mapping from a standard normal distribution in a low-dimensional latent space into a higher-dimensional data space. As such, these mappings are only capable of producing simple data topologies, i.e., those equivalent to an immersion of Euclidean space. In this work, we demonstrate the limitations of such latent space generative models when trained on data distributions with non-trivial topologies. We do this by training these models on synthetic image datasets with known topologies (spheres, torii, etc.). We then show how this results in failures of both data generation as well as data interpolation. Next, we compare this behavior to two classes of deep generative models that in principle allow for more complex data topologies. First, we look at chart autoencoders (CAEs), which construct a smooth data manifold from multiple latent space chart mappings. Second, we explore score-based models, e.g., denoising diffusion probabilistic models, which estimate gradients of the data distribution without resorting to an explicit mapping to a latent space. Our results show that these models do demonstrate improved ability over latent space models in modeling data distributions with complex topologies, however, challenges still remain.<\/jats:p>","DOI":"10.3389\/fcomp.2024.1260604","type":"journal-article","created":{"date-parts":[[2024,8,26]],"date-time":"2024-08-26T05:04:42Z","timestamp":1724648682000},"update-policy":"https:\/\/doi.org\/10.3389\/crossmark-policy","source":"Crossref","is-referenced-by-count":3,"title":["Implications of data topology for deep generative models"],"prefix":"10.3389","volume":"6","author":[{"given":"Yinzhu","family":"Jin","sequence":"first","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Rory","family":"McDaniel","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"N. Joseph","family":"Tatro","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Michael J.","family":"Catanzaro","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Abraham D.","family":"Smith","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Paul","family":"Bendich","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Matthew B.","family":"Dwyer","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"P. Thomas","family":"Fletcher","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]}],"member":"1965","published-online":{"date-parts":[[2024,8,26]]},"reference":[{"key":"B1","article-title":"\u201cLatent space oddity: on the curvature of deep generative models,\u201d","author":"Arvanitidis","year":"2018","journal-title":"Proceedings of the 6th International Conference on Learning Representations, ICLR 2018"},{"key":"B2","first-page":"7294","article-title":"\u201cManifold topology divergence: a framework for comparing data manifolds,\u201d","author":"Barannikov","year":"2021","journal-title":"Proceedings of Advances in Neural Information Processing Systems 34: Annual Conference on Neural Information Processing Systems 2021, NeurIPS 2021"},{"key":"B3","doi-asserted-by":"publisher","first-page":"103329","DOI":"10.1016\/j.cviu.2021.103329","article-title":"Pros and cons of GAN evaluation measures: new developments","volume":"215","author":"Borji","year":"2022","journal-title":"Comput. Vis. Image Underst"},{"key":"B4","first-page":"2786","article-title":"\u201cPerslay: a neural network layer for persistence diagrams and new graph topological signatures,\u201d","author":"Carri\u00e9re","year":"2020","journal-title":"Proceedings of the 23rd International Conference on Artificial Intelligence and Statistics, AISTATS 2020"},{"key":"B5","first-page":"2573","article-title":"\u201cA topological regularizer for classifiers via persistent homology,\u201d","author":"Chen","year":"2019","journal-title":"Proceedings of the 22nd International Conference on Artificial Intelligence and Statistics, AISTATS 2019"},{"key":"B6","first-page":"1540","article-title":"\u201cMetrics for deep generative models,\u201d","author":"Chen","year":"2018","journal-title":"Proceedings of the 21st International Conference on Artificial Intelligence and Statistics, AISTATS 2018"},{"key":"B7","article-title":"Effectively unbiased FID and inception score and where to find them","author":"Chong","year":"2020","journal-title":"arXiv:1911.07023"},{"key":"B8","doi-asserted-by":"publisher","first-page":"103","DOI":"10.1007\/s00454-006-1276-5","article-title":"Stability of persistence diagrams","volume":"37","author":"Cohen-Steiner","year":"2007","journal-title":"Discr. Comput. Geom"},{"key":"B9","article-title":"Implicit generation and generalization in energy-based models","author":"Du","year":"2019","journal-title":"arXiv:1903.08689"},{"key":"B10","doi-asserted-by":"publisher","DOI":"10.1090\/mbk\/069","author":"Edelsbrunner","year":"2010","journal-title":"Computational Topology"},{"key":"B11","first-page":"1436","article-title":"\u201cBOHB: robust and efficient hyperparameter optimization at scale,\u201d","author":"Falkner","year":"2018","journal-title":"Proceedings of the 35th International Conference on Machine Learning, ICML 2018"},{"key":"B12","doi-asserted-by":"publisher","first-page":"2301","DOI":"10.1214\/14-AOS1252","article-title":"Confidence sets for persistence diagrams","volume":"42","author":"Fasy","year":"2014","journal-title":"Ann. Stat"},{"key":"B13","first-page":"1608","article-title":"\u201cLearning generative models with sinkhorn divergences,\u201d","author":"Genevay","year":"2018","journal-title":"International Conference on Artificial Intelligence and Statistics, AISTATS 2018"},{"key":"B14","first-page":"2672","article-title":"\u201cGenerative adversarial nets,\u201d","author":"Goodfellow","year":"2014","journal-title":"Proceeding of Advances in Neural Information Processing Systems 27: Annual Conference on Neural Information Processing Systems 2014, NEURIPS 2014"},{"key":"B15","doi-asserted-by":"publisher","first-page":"7036","DOI":"10.1021\/ja01002a036","article-title":"Molecular geometry. V. Evaluation of functions and conformations of medium rings","volume":"89","author":"Hendrickson","year":"1967","journal-title":"J. Am. Chem. Soc"},{"key":"B16","doi-asserted-by":"publisher","first-page":"681108","DOI":"10.3389\/frai.2021.681108","article-title":"A survey of topological machine learning methods","volume":"4","author":"Hensel","year":"2021","journal-title":"Front. Artif. Intell"},{"key":"B17","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":"B18","doi-asserted-by":"publisher","first-page":"504","DOI":"10.1126\/science.1127647","article-title":"Reducing the dimensionality of data with neural networks","volume":"313","author":"Hinton","year":"2006","journal-title":"Science"},{"key":"B19","doi-asserted-by":"publisher","first-page":"6840","DOI":"10.48550\/arXiv.2006.11239","article-title":"Denoising diffusion probabilistic models","volume":"33","author":"Ho","year":"2020","journal-title":"Adv. Neural Inf. Process. Syst"},{"key":"B20","doi-asserted-by":"publisher","DOI":"10.1007\/978-3-642-73778-7_164","article-title":"\u201cA shortest augmenting path algorithm for dense and sparse linear assignment problems,\u201d","author":"Jonker","year":"1988","journal-title":"Proceedings of the 16th Annual Meeting of DGOR in Cooperation with NSOR"},{"key":"B21","first-page":"2626","article-title":"\u201cGeometry score: a method for comparing generative adversarial networks,\u201d","author":"Khrulkov","year":"2018","journal-title":"Proceedings of the 35th International Conference on Machine Learning, ICML 2018"},{"key":"B22","article-title":"\u201cAdam: a method for stochastic optimization,\u201d","author":"Kingma","year":"2015","journal-title":"3rd International Conference on Learning Representations, ICLR 2015"},{"key":"B23","article-title":"\u201cAuto-encoding variational bayes,\u201d","author":"Kingma","year":"2014","journal-title":"Proceedings of the 2nd International Conference on Learning Representations, ICLR 2014"},{"key":"B24","doi-asserted-by":"publisher","first-page":"234115","DOI":"10.1063\/1.3445267","article-title":"Topology of cyclo-octane energy landscape","volume":"132","author":"Martin","year":"2010","journal-title":"J. Chem. Phys"},{"key":"B25","first-page":"7176","article-title":"\u201cReliable fidelity and diversity metrics for generative models,\u201d","author":"Naeem","year":"2020","journal-title":"Proceedings of the 37th International Conference on Machine Learning, ICML 2020"},{"key":"B26","doi-asserted-by":"publisher","first-page":"7503","DOI":"10.48550\/arXiv.2004.06093","article-title":"Topology of deep neural networks","volume":"184","author":"Naitzat","year":"2020","journal-title":"J. Mach. Learn. Res"},{"key":"B27","article-title":"Topological optimization with big steps","author":"Nigmetov","year":"2022","journal-title":"arXiv:2203.16748"},{"key":"B28","doi-asserted-by":"publisher","first-page":"1","DOI":"10.1140\/epjds\/s13688-017-0109-5","article-title":"A roadmap for the computation of persistent homology","volume":"6","author":"Otter","year":"2017","journal-title":"EPJ Data Sci"},{"key":"B29","volume-title":"Persistence Theory: From Quiver Representations to Data Analysis","author":"Oudot","year":"2017"},{"key":"B30","article-title":"On aliased resizing and surprising subtleties in GAN evaluation","author":"Parmar","year":"2022","journal-title":"arXiv:2104.11222"},{"key":"B31","unstructured":"Persistence of vision raytracer (version 3.6)2004"},{"key":"B32","first-page":"1530","article-title":"\u201cVariational inference with normalizing flows,\u201d","author":"Rezende","year":"2015","journal-title":"Proceedings of the 32nd International Conference on Machine Learning, ICML 2015"},{"key":"B33","first-page":"5234","article-title":"\u201cAssessing generative models via precision and recall,\u201d","author":"Sajjadi","year":"2018","journal-title":"Proceedings of Advances in Neural Information Processing Systems 31: Annual Conference on Neural Information Processing Systems 2018, NeurIPS 2018"},{"key":"B34","article-title":"Chart auto-encoders for manifold structured data","author":"Schonsheck","year":"2019","journal-title":"arXiv:1912.10094"},{"key":"B35","doi-asserted-by":"publisher","DOI":"10.1109\/CVPRW.2018.00071","article-title":"\u201cThe Riemannian geometry of deep generative models,\u201d","author":"Shao","year":"2018","journal-title":"Proceedings of 2018 IEEE Conference on Computer Vision and Pattern Recognition Workshops, CVPR Workshops 2018"},{"key":"B36","doi-asserted-by":"publisher","first-page":"1","DOI":"10.1145\/3293353.3293422","article-title":"\u201cGeometry of deep generative models for disentangled representations,\u201d","volume":"68","author":"Shukla","year":"2018","journal-title":"Proceedings of ICVGIP 2018: 11th Indian Conference on Computer Vision, Graphics and Image Processing"},{"key":"B37","first-page":"2256","article-title":"\u201cDeep unsupervised learning using nonequilibrium thermodynamics,\u201d","author":"Sohl-Dickstein","year":"2015","journal-title":"Proceedings of the 32nd International Conference on Machine Learning, ICML 2015"},{"key":"B38","first-page":"109","article-title":"\u201cA fast and robust method for global topological functional optimization,\u201d","author":"Solomon","year":"2021","journal-title":"Proceedings of the 24th International Conference on Artificial Intelligence and Statistics, AISTATS 2021"},{"key":"B39","article-title":"\u201cScore-based generative modeling through stochastic differential equations,\u201d","author":"Song","year":"2021","journal-title":"Proceedings of the 9th International Conference on Learning Representations, ICLR"},{"key":"B40","doi-asserted-by":"publisher","first-page":"1","DOI":"10.48550\/arXiv.2004.02551","article-title":"giotto-tda:: a topological data analysis toolkit for machine learning and data exploration","volume":"39","author":"Tauzin","year":"2021","journal-title":"J. Mach. Learn. Res"},{"key":"B41","doi-asserted-by":"publisher","first-page":"261","DOI":"10.1038\/s41592-019-0686-2","article-title":"SciPy 1.0: fundamental algorithms for scientific computing in python","volume":"17","author":"Virtanen","year":"2020","journal-title":"Nat. Methods"},{"key":"B42","doi-asserted-by":"publisher","DOI":"10.1109\/BigData52589.2021.9671368","article-title":"\u201cActivation landscapes as a topological summary of neural network performance,\u201d","author":"Wheeler","year":"2021","journal-title":"Proceedings of 2021 IEEE International Conference on Big Data"}],"container-title":["Frontiers in Computer Science"],"original-title":[],"link":[{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/fcomp.2024.1260604\/full","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2024,8,26]],"date-time":"2024-08-26T05:05:09Z","timestamp":1724648709000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/fcomp.2024.1260604\/full"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2024,8,26]]},"references-count":42,"alternative-id":["10.3389\/fcomp.2024.1260604"],"URL":"https:\/\/doi.org\/10.3389\/fcomp.2024.1260604","relation":{},"ISSN":["2624-9898"],"issn-type":[{"value":"2624-9898","type":"electronic"}],"subject":[],"published":{"date-parts":[[2024,8,26]]},"article-number":"1260604"}}