{"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/feedback-networks","title":"Feedback Networks","arxiv_id":"1612.09508","date":"2016-12-30","proceeding":"CVPR 2017 7","authors":["Amir R. Zamir","Te-Lin Wu","Lin Sun","William Shen","Jitendra Malik","Silvio Savarese"],"abstract":"Currently, the most successful learning models in computer vision are based\non learning successive representations followed by a decision layer. This is\nusually actualized through feedforward multilayer neural networks, e.g.\nConvNets, where each layer forms one of such successive representations.\nHowever, an alternative that can achieve the same goal is a feedback based\napproach in which the representation is formed in an iterative manner based on\na feedback received from previous iteration's output.\n  We establish that a feedback based approach has several fundamental\nadvantages over feedforward: it enables making early predictions at the query\ntime, its output naturally conforms to a hierarchical structure in the label\nspace (e.g. a taxonomy), and it provides a new basis for Curriculum Learning.\nWe observe that feedback networks develop a considerably different\nrepresentation compared to feedforward counterparts, in line with the\naforementioned advantages. We put forth a general feedback based learning\narchitecture with the endpoint results on par or better than existing\nfeedforward networks with the addition of the above advantages. We also\ninvestigate several mechanisms in feedback architectures (e.g. skip connections\nin time) and design choices (e.g. feedback length). We hope this study offers\nnew perspectives in quest for more natural and practical learning models.","url_abs":"http://arxiv.org/abs/1612.09508v3","url_pdf":"http://arxiv.org/pdf/1612.09508v3.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":"feedback-networks","repo_url":"https://github.com/StanfordVL/feedback-networks","is_official":1,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"torch","reach":{"status":"ok","spdx":"MIT"}}],"tasks":[],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1612.09508","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}