{"oa":1,"oa_version":"Preprint","user_id":"2DF688A6-F248-11E8-B48F-1D18A9856A87","date_created":"2024-02-07T14:28:34Z","year":"2023","title":"Multi-view causal representation learning with partial observability","acknowledgement":"This work was initiated at the Second Bellairs Workshop on Causality held at the Bellairs Research Institute, January 6–13, 2022; we thank all workshop participants for providing a stimulating research environment. Further, we thank Cian Eastwood, Luigi Gresele, Stefano Soatto, Marco Bagatella, and A. René Geist for helpful discussion. GM is a member of the Machine Learning Cluster of Excellence, EXC number 2064/1 – Project number 390727645. JvK and GM acknowledge support from the German Federal Ministry of Education and Research (BMBF) through the Tübingen AI Center (FKZ: 01IS18039B). The research of DX and SM was supported by the Air Force Office of Scientific Research under award number FA8655-22-1-7155. Any opinions, findings, and conclusions or recommendations expressed in\r\nthis material are those of the author(s) and do not necessarily reflect the views of the United States Air Force. We also thank SURF for the support in using the Dutch National Supercomputer Snellius. DY was supported by an Amazon fellowship and the International Max Planck Research School for Intelligent Systems (IMPRS-IS). Work done outside of Amazon. SL was supported by an IVADO excellence PhD scholarship and by Samsung Electronics Co., Ldt.","status":"public","type":"preprint","doi":"10.48550/arXiv.2311.04056","day":"07","author":[{"last_name":"Yao","first_name":"Dingling","full_name":"Yao, Dingling","id":"d3e02e50-48a8-11ee-8f62-c108061797fa"},{"full_name":"Xu, Danru","last_name":"Xu","first_name":"Danru"},{"last_name":"Lachapelle","first_name":"Sébastien","full_name":"Lachapelle, Sébastien"},{"first_name":"Sara","last_name":"Magliacane","full_name":"Magliacane, Sara"},{"full_name":"Taslakian, Perouz","first_name":"Perouz","last_name":"Taslakian"},{"first_name":"Georg","last_name":"Martius","full_name":"Martius, Georg"},{"full_name":"Kügelgen, Julius von","last_name":"Kügelgen","first_name":"Julius von"},{"orcid":"0000-0002-4850-0683","id":"26cfd52f-2483-11ee-8040-88983bcc06d4","full_name":"Locatello, Francesco","last_name":"Locatello","first_name":"Francesco"}],"publication":"arXiv","external_id":{"arxiv":["2311.04056"]},"month":"11","article_processing_charge":"No","date_updated":"2024-02-12T08:07:33Z","date_published":"2023-11-07T00:00:00Z","publication_status":"submitted","citation":{"ieee":"D. Yao <i>et al.</i>, “Multi-view causal representation learning with partial observability,” <i>arXiv</i>. .","mla":"Yao, Dingling, et al. “Multi-View Causal Representation Learning with Partial Observability.” <i>ArXiv</i>, 2311.04056, doi:<a href=\"https://doi.org/10.48550/arXiv.2311.04056\">10.48550/arXiv.2311.04056</a>.","short":"D. Yao, D. Xu, S. Lachapelle, S. Magliacane, P. Taslakian, G. Martius, J. von Kügelgen, F. Locatello, ArXiv (n.d.).","ista":"Yao D, Xu D, Lachapelle S, Magliacane S, Taslakian P, Martius G, Kügelgen J von, Locatello F. Multi-view causal representation learning with partial observability. arXiv, 2311.04056.","chicago":"Yao, Dingling, Danru Xu, Sébastien Lachapelle, Sara Magliacane, Perouz Taslakian, Georg Martius, Julius von Kügelgen, and Francesco Locatello. “Multi-View Causal Representation Learning with Partial Observability.” <i>ArXiv</i>, n.d. <a href=\"https://doi.org/10.48550/arXiv.2311.04056\">https://doi.org/10.48550/arXiv.2311.04056</a>.","apa":"Yao, D., Xu, D., Lachapelle, S., Magliacane, S., Taslakian, P., Martius, G., … Locatello, F. (n.d.). Multi-view causal representation learning with partial observability. <i>arXiv</i>. <a href=\"https://doi.org/10.48550/arXiv.2311.04056\">https://doi.org/10.48550/arXiv.2311.04056</a>","ama":"Yao D, Xu D, Lachapelle S, et al. Multi-view causal representation learning with partial observability. <i>arXiv</i>. doi:<a href=\"https://doi.org/10.48550/arXiv.2311.04056\">10.48550/arXiv.2311.04056</a>"},"department":[{"_id":"FrLo"}],"language":[{"iso":"eng"}],"article_number":"2311.04056","_id":"14946","abstract":[{"text":"We present a unified framework for studying the identifiability of\r\nrepresentations learned from simultaneously observed views, such as different\r\ndata modalities. We allow a partially observed setting in which each view\r\nconstitutes a nonlinear mixture of a subset of underlying latent variables,\r\nwhich can be causally related. We prove that the information shared across all\r\nsubsets of any number of views can be learned up to a smooth bijection using\r\ncontrastive learning and a single encoder per view. We also provide graphical\r\ncriteria indicating which latent variables can be identified through a simple\r\nset of rules, which we refer to as identifiability algebra. Our general\r\nframework and theoretical results unify and extend several previous works on\r\nmulti-view nonlinear ICA, disentanglement, and causal representation learning.\r\nWe experimentally validate our claims on numerical, image, and multi-modal data\r\nsets. Further, we demonstrate that the performance of prior methods is\r\nrecovered in different special cases of our setup. Overall, we find that access\r\nto multiple partial views enables us to identify a more fine-grained\r\nrepresentation, under the generally milder assumption of partial observability.","lang":"eng"}],"main_file_link":[{"url":"https://doi.org/10.48550/arXiv.2311.04056","open_access":"1"}]}