{"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/a-theory-of-generative-convnet","title":"A Theory of Generative ConvNet","arxiv_id":"1602.03264","date":"2016-02-10","proceeding":null,"authors":["Jianwen Xie","Yang Lu","Song-Chun Zhu","Ying Nian Wu"],"abstract":"We show that a generative random field model, which we call generative\nConvNet, can be derived from the commonly used discriminative ConvNet, by\nassuming a ConvNet for multi-category classification and assuming one of the\ncategories is a base category generated by a reference distribution. If we\nfurther assume that the non-linearity in the ConvNet is Rectified Linear Unit\n(ReLU) and the reference distribution is Gaussian white noise, then we obtain a\ngenerative ConvNet model that is unique among energy-based models: The model is\npiecewise Gaussian, and the means of the Gaussian pieces are defined by an\nauto-encoder, where the filters in the bottom-up encoding become the basis\nfunctions in the top-down decoding, and the binary activation variables\ndetected by the filters in the bottom-up convolution process become the\ncoefficients of the basis functions in the top-down deconvolution process. The\nLangevin dynamics for sampling the generative ConvNet is driven by the\nreconstruction error of this auto-encoder. The contrastive divergence learning\nof the generative ConvNet reconstructs the training images by the auto-encoder.\nThe maximum likelihood learning algorithm can synthesize realistic natural\nimage patterns.","url_abs":"http://arxiv.org/abs/1602.03264v3","url_pdf":"http://arxiv.org/pdf/1602.03264v3.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":[],"tasks":[],"methods":[{"method_slug":"convolution","method_name":"Convolution"},{"method_slug":"ebm","method_name":"EBM"}],"datasets_introduced":[],"methods_introduced":[{"slug":"ebm","name":"EBM","full_name":"energy-based model"}],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1602.03264","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}