{"year":"2020","abstract":[{"lang":"eng","text":"The idea behind the unsupervised learning of disentangled representations is that real-world data is generated by a few explanatory factors of variation which can be recovered by unsupervised learning algorithms. In this paper, we provide a sober look at recent progress in the field and challenge some common assumptions. We first theoretically show that the unsupervised learning of disentangled representations is fundamentally impossible without inductive biases on both the models and the data. Then, we train over 14000\r\n models covering most prominent methods and evaluation metrics in a reproducible large-scale experimental study on eight data sets. We observe that while the different methods successfully enforce properties “encouraged” by the corresponding losses, well-disentangled models seemingly cannot be identified without supervision. Furthermore, different evaluation metrics do not always agree on what should be considered “disentangled” and exhibit systematic differences in the estimation. Finally, increased disentanglement does not seem to necessarily lead to a decreased sample complexity of learning for downstream tasks. Our results suggest that future work on disentanglement learning should be explicit about the role of inductive biases and (implicit) supervision, investigate concrete benefits of enforcing disentanglement of the learned representations, and consider a reproducible experimental setup covering several data sets."}],"oa":1,"main_file_link":[{"open_access":"1","url":"https://jmlr.csail.mit.edu/papers/v21/19-976.html"}],"author":[{"last_name":"Locatello","orcid":"0000-0002-4850-0683","first_name":"Francesco","id":"26cfd52f-2483-11ee-8040-88983bcc06d4","full_name":"Locatello, Francesco"},{"full_name":"Bauer, Stefan","last_name":"Bauer","first_name":"Stefan"},{"full_name":"Lucic, Mario","last_name":"Lucic","first_name":"Mario"},{"first_name":"Gunnar","last_name":"Rätsch","full_name":"Rätsch, Gunnar"},{"full_name":"Gelly, Sylvain","last_name":"Gelly","first_name":"Sylvain"},{"last_name":"Schölkopf","first_name":"Bernhard","full_name":"Schölkopf, Bernhard"},{"full_name":"Bachem, Olivier","last_name":"Bachem","first_name":"Olivier"}],"day":"01","language":[{"iso":"eng"}],"article_number":"209","tmp":{"name":"Creative Commons Attribution 4.0 International Public License (CC-BY 4.0)","legal_code_url":"https://creativecommons.org/licenses/by/4.0/legalcode","short":"CC BY (4.0)","image":"/images/cc_by.png"},"has_accepted_license":"1","ddc":["000"],"title":"A sober look at the unsupervised learning of disentangled representations and their evaluation","intvolume":"        21","citation":{"ieee":"F. Locatello <i>et al.</i>, “A sober look at the unsupervised learning of disentangled representations and their evaluation,” <i>Journal of Machine Learning Research</i>, vol. 21. MIT Press, 2020.","ama":"Locatello F, Bauer S, Lucic M, et al. A sober look at the unsupervised learning of disentangled representations and their evaluation. <i>Journal of Machine Learning Research</i>. 2020;21.","chicago":"Locatello, Francesco, Stefan Bauer, Mario Lucic, Gunnar Rätsch, Sylvain Gelly, Bernhard Schölkopf, and Olivier Bachem. “A Sober Look at the Unsupervised Learning of Disentangled Representations and Their Evaluation.” <i>Journal of Machine Learning Research</i>. MIT Press, 2020.","apa":"Locatello, F., Bauer, S., Lucic, M., Rätsch, G., Gelly, S., Schölkopf, B., &#38; Bachem, O. (2020). A sober look at the unsupervised learning of disentangled representations and their evaluation. <i>Journal of Machine Learning Research</i>. MIT Press.","short":"F. Locatello, S. Bauer, M. Lucic, G. Rätsch, S. Gelly, B. Schölkopf, O. Bachem, Journal of Machine Learning Research 21 (2020).","mla":"Locatello, Francesco, et al. “A Sober Look at the Unsupervised Learning of Disentangled Representations and Their Evaluation.” <i>Journal of Machine Learning Research</i>, vol. 21, 209, MIT Press, 2020.","ista":"Locatello F, Bauer S, Lucic M, Rätsch G, Gelly S, Schölkopf B, Bachem O. 2020. A sober look at the unsupervised learning of disentangled representations and their evaluation. Journal of Machine Learning Research. 21, 209."},"article_processing_charge":"No","scopus_import":"1","date_published":"2020-09-01T00:00:00Z","volume":21,"quality_controlled":"1","external_id":{"arxiv":["2010.14766"]},"publisher":"MIT Press","license":"https://creativecommons.org/licenses/by/4.0/","publication_status":"published","date_created":"2023-08-22T14:10:34Z","month":"09","publication":"Journal of Machine Learning Research","type":"journal_article","date_updated":"2023-09-12T09:23:56Z","user_id":"2DF688A6-F248-11E8-B48F-1D18A9856A87","department":[{"_id":"FrLo"}],"oa_version":"Published Version","extern":"1","_id":"14195","article_type":"original","status":"public"}