{"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/learning-discrete-representations-via","title":"Learning Discrete Representations via Information Maximizing Self-Augmented Training","arxiv_id":"1702.08720","date":"2017-02-28","proceeding":"ICML 2017 8","authors":["Weihua Hu","Takeru Miyato","Seiya Tokui","Eiichi Matsumoto","Masashi Sugiyama"],"abstract":"Learning discrete representations of data is a central machine learning task\nbecause of the compactness of the representations and ease of interpretation.\nThe task includes clustering and hash learning as special cases. Deep neural\nnetworks are promising to be used because they can model the non-linearity of\ndata and scale to large datasets. However, their model complexity is huge, and\ntherefore, we need to carefully regularize the networks in order to learn\nuseful representations that exhibit intended invariance for applications of\ninterest. To this end, we propose a method called Information Maximizing\nSelf-Augmented Training (IMSAT). In IMSAT, we use data augmentation to impose\nthe invariance on discrete representations. More specifically, we encourage the\npredicted representations of augmented data points to be close to those of the\noriginal data points in an end-to-end fashion. At the same time, we maximize\nthe information-theoretic dependency between data and their predicted discrete\nrepresentations. Extensive experiments on benchmark datasets show that IMSAT\nproduces state-of-the-art results for both clustering and unsupervised hash\nlearning.","url_abs":"http://arxiv.org/abs/1702.08720v3","url_pdf":"http://arxiv.org/pdf/1702.08720v3.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":"learning-discrete-representations-via","repo_url":"https://github.com/weihua916/imsat","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"none","reach":{"status":"ok"}},{"paper_slug":"learning-discrete-representations-via","repo_url":"https://github.com/MOhammedJAbi/Imsat","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}}],"tasks":[{"task_slug":"clustering","task_name":"Clustering"},{"task_slug":"data-augmentation","task_name":"Data Augmentation"},{"task_slug":"unsupervised-image-classification","task_name":"Unsupervised Image Classification"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/unsupervised-image-classification-on-svhn","task":"Unsupervised Image Classification","dataset":"SVHN","model":"IMSAT","rank_in_archive_order":3,"of":4,"metrics":{"# of clusters (k)":"10","Acc":"57.30"},"uses_additional_data":true}],"syntology":{"syntology_url":null,"atlas_url":"https://app.syntology.ai/?focus=1702.08720","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}