{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/unsupervised-feature-extraction-by-time","title":"Unsupervised Feature Extraction by Time-Contrastive Learning and Nonlinear ICA","arxiv_id":"1605.06336","date":"2016-05-20","proceeding":"NeurIPS 2016 12","authors":["Aapo Hyvarinen","Hiroshi Morioka"],"abstract":"Nonlinear independent component analysis (ICA) provides an appealing\nframework for unsupervised feature learning, but the models proposed so far are\nnot identifiable. Here, we first propose a new intuitive principle of\nunsupervised deep learning from time series which uses the nonstationary\nstructure of the data. Our learning principle, time-contrastive learning (TCL),\nfinds a representation which allows optimal discrimination of time segments\n(windows). Surprisingly, we show how TCL can be related to a nonlinear ICA\nmodel, when ICA is redefined to include temporal nonstationarities. In\nparticular, we show that TCL combined with linear ICA estimates the nonlinear\nICA model up to point-wise transformations of the sources, and this solution is\nunique --- thus providing the first identifiability result for nonlinear ICA\nwhich is rigorous, constructive, as well as very general.","url_abs":"http://arxiv.org/abs/1605.06336v1","url_pdf":"http://arxiv.org/pdf/1605.06336v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"unsupervised-feature-extraction-by-time","repo_url":"https://github.com/ilkhem/icebeem","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null},{"paper_slug":"unsupervised-feature-extraction-by-time","repo_url":"https://github.com/kondratevakate/fmri-component-analysis","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}}],"tasks":[{"task_slug":"contrastive-learning","task_name":"Contrastive Learning"},{"task_slug":"time-series-1","task_name":"Time Series"},{"task_slug":"time-series","task_name":"Time Series Analysis"}],"methods":[{"method_slug":"ica","method_name":"ICA"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1605.06336","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}