{"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/disentangling-factors-of-variation-in-deep","title":"Disentangling factors of variation in deep representations using adversarial training","arxiv_id":"1611.03383","date":"2016-11-10","proceeding":null,"authors":["Michael Mathieu","Junbo Zhao","Pablo Sprechmann","Aditya Ramesh","Yann Lecun"],"abstract":"We introduce a conditional generative model for learning to disentangle the\nhidden factors of variation within a set of labeled observations, and separate\nthem into complementary codes. One code summarizes the specified factors of\nvariation associated with the labels. The other summarizes the remaining\nunspecified variability. During training, the only available source of\nsupervision comes from our ability to distinguish among different observations\nbelonging to the same class. Examples of such observations include images of a\nset of labeled objects captured at different viewpoints, or recordings of set\nof speakers dictating multiple phrases. In both instances, the intra-class\ndiversity is the source of the unspecified factors of variation: each object is\nobserved at multiple viewpoints, and each speaker dictates multiple phrases.\nLearning to disentangle the specified factors from the unspecified ones becomes\neasier when strong supervision is possible. Suppose that during training, we\nhave access to pairs of images, where each pair shows two different objects\ncaptured from the same viewpoint. This source of alignment allows us to solve\nour task using existing methods. However, labels for the unspecified factors\nare usually unavailable in realistic scenarios where data acquisition is not\nstrictly controlled. We address the problem of disentanglement in this more\ngeneral setting by combining deep convolutional autoencoders with a form of\nadversarial training. Both factors of variation are implicitly captured in the\norganization of the learned embedding space, and can be used for solving\nsingle-image analogies. Experimental results on synthetic and real datasets\nshow that the proposed method is capable of generalizing to unseen classes and\nintra-class variabilities.","url_abs":"http://arxiv.org/abs/1611.03383v1","url_pdf":"http://arxiv.org/pdf/1611.03383v1.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":"disentangling-factors-of-variation-in-deep","repo_url":"https://github.com/ananyahjha93/challenges-in-disentangling","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}},{"paper_slug":"disentangling-factors-of-variation-in-deep","repo_url":"https://github.com/ananyahjha93/cycle-consistent-vae","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}},{"paper_slug":"disentangling-factors-of-variation-in-deep","repo_url":"https://github.com/ananyahjha93/disentangling-factors-of-variation-using-adversarial-training","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}}],"tasks":[{"task_slug":"disentanglement","task_name":"Disentanglement"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1611.03383","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}