{"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/max-margin-deep-generative-models-for-semi","title":"Max-Margin Deep Generative Models for (Semi-)Supervised Learning","arxiv_id":"1611.07119","date":"2016-11-22","proceeding":null,"authors":["Chongxuan Li","Jun Zhu","Bo Zhang"],"abstract":"Deep generative models (DGMs) are effective on learning multilayered\nrepresentations of complex data and performing inference of input data by\nexploring the generative ability. However, it is relatively insufficient to\nempower the discriminative ability of DGMs on making accurate predictions. This\npaper presents max-margin deep generative models (mmDGMs) and a\nclass-conditional variant (mmDCGMs), which explore the strongly discriminative\nprinciple of max-margin learning to improve the predictive performance of DGMs\nin both supervised and semi-supervised learning, while retaining the generative\ncapability. In semi-supervised learning, we use the predictions of a max-margin\nclassifier as the missing labels instead of performing full posterior inference\nfor efficiency; we also introduce additional max-margin and label-balance\nregularization terms of unlabeled data for effectiveness. We develop an\nefficient doubly stochastic subgradient algorithm for the piecewise linear\nobjectives in different settings. Empirical results on various datasets\ndemonstrate that: (1) max-margin learning can significantly improve the\nprediction performance of DGMs and meanwhile retain the generative ability; (2)\nin supervised learning, mmDGMs are competitive to the best fully discriminative\nnetworks when employing convolutional neural networks as the generative and\nrecognition models; and (3) in semi-supervised learning, mmDCGMs can perform\nefficient inference and achieve state-of-the-art classification results on\nseveral benchmarks.","url_abs":"http://arxiv.org/abs/1611.07119v1","url_pdf":"http://arxiv.org/pdf/1611.07119v1.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":"max-margin-deep-generative-models-for-semi","repo_url":"https://github.com/thu-ml/mmdcgm-ssl","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"none","reach":{"status":"ok"}}],"tasks":[{"task_slug":"missing-labels","task_name":"Missing Labels"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1611.07119","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}