{"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/rocket-launching-a-universal-and-efficient","title":"Rocket Launching: A Universal and Efficient Framework for Training Well-performing Light Net","arxiv_id":"1708.04106","date":"2017-08-14","proceeding":null,"authors":["Guorui Zhou","Ying Fan","Runpeng Cui","Weijie Bian","Xiaoqiang Zhu","Kun Gai"],"abstract":"Models applied on real time response task, like click-through rate (CTR)\nprediction model, require high accuracy and rigorous response time. Therefore,\ntop-performing deep models of high depth and complexity are not well suited for\nthese applications with the limitations on the inference time. In order to\nfurther improve the neural networks' performance given the time and\ncomputational limitations, we propose an approach that exploits a cumbersome\nnet to help train the lightweight net for prediction. We dub the whole process\nrocket launching, where the cumbersome booster net is used to guide the\nlearning of the target light net throughout the whole training process. We\nanalyze different loss functions aiming at pushing the light net to behave\nsimilarly to the booster net, and adopt the loss with best performance in our\nexperiments. We use one technique called gradient block to improve the\nperformance of the light net and booster net further. Experiments on benchmark\ndatasets and real-life industrial advertisement data present that our light\nmodel can get performance only previously achievable with more complex models.","url_abs":"http://arxiv.org/abs/1708.04106v3","url_pdf":"http://arxiv.org/pdf/1708.04106v3.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":"rocket-launching-a-universal-and-efficient","repo_url":"https://github.com/zhougr1993/Rocket-Launching","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}},{"paper_slug":"rocket-launching-a-universal-and-efficient","repo_url":"https://github.com/alibaba/EasyRec/blob/master/easy_rec/python/model/rocket_launching.py","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"tf","reach":null},{"paper_slug":"rocket-launching-a-universal-and-efficient","repo_url":"https://github.com/alibaba/TorchEasyRec","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"click-through-rate-prediction","task_name":"Click-Through Rate Prediction"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":"https://syntology.ai/paper/1708.04106","atlas_url":"https://app.syntology.ai/?focus=1708.04106","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}