{"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/splitnet-divide-and-co-training","title":"Towards Better Accuracy-efficiency Trade-offs: Divide and Co-training","arxiv_id":"2011.14660","date":"2020-11-30","proceeding":null,"authors":["Shuai Zhao","Liguang Zhou","Wenxiao Wang","Deng Cai","Tin Lun Lam","Yangsheng Xu"],"abstract":"The width of a neural network matters since increasing the width will necessarily increase the model capacity. However, the performance of a network does not improve linearly with the width and soon gets saturated. In this case, we argue that increasing the number of networks (ensemble) can achieve better accuracy-efficiency trade-offs than purely increasing the width. To prove it, one large network is divided into several small ones regarding its parameters and regularization components. Each of these small networks has a fraction of the original one's parameters. We then train these small networks together and make them see various views of the same data to increase their diversity. During this co-training process, networks can also learn from each other. As a result, small networks can achieve better ensemble performance than the large one with few or no extra parameters or FLOPs, \\ie, achieving better accuracy-efficiency trade-offs. Small networks can also achieve faster inference speed than the large one by concurrent running. All of the above shows that the number of networks is a new dimension of model scaling. We validate our argument with 8 different neural architectures on common benchmarks through extensive experiments. The code is available at \\url{https://github.com/FreeformRobotics/Divide-and-Co-training}.","url_abs":"https://arxiv.org/abs/2011.14660v4","url_pdf":"https://arxiv.org/pdf/2011.14660v4.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":"splitnet-divide-and-co-training","repo_url":"https://github.com/freeformrobotics/divide-and-co-training","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}},{"paper_slug":"splitnet-divide-and-co-training","repo_url":"https://github.com/mzhaoshuai/Divide-and-Co-training","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}}],"tasks":[{"task_slug":"image-classification","task_name":"Image Classification"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/image-classification-on-cifar-10","task":"Image Classification","dataset":"CIFAR-10","model":"PyramidNet-272, S=4","rank_in_archive_order":30,"of":265,"metrics":{"Percentage correct":"98.71"},"uses_additional_data":false},{"leaderboard":"/sota/image-classification-on-cifar-10","task":"Image Classification","dataset":"CIFAR-10","model":"WRN-40-10, S=4","rank_in_archive_order":42,"of":265,"metrics":{"Percentage correct":"98.38"},"uses_additional_data":false},{"leaderboard":"/sota/image-classification-on-cifar-10","task":"Image Classification","dataset":"CIFAR-10","model":"WRN-28-10, S=4","rank_in_archive_order":43,"of":265,"metrics":{"Percentage correct":"98.32"},"uses_additional_data":false},{"leaderboard":"/sota/image-classification-on-cifar-10","task":"Image Classification","dataset":"CIFAR-10","model":"Shake-Shake 26 2x96d, S=4","rank_in_archive_order":44,"of":265,"metrics":{"Percentage correct":"98.31"},"uses_additional_data":false},{"leaderboard":"/sota/image-classification-on-cifar-100","task":"Image Classification","dataset":"CIFAR-100","model":"PyramidNet-272, S=4","rank_in_archive_order":30,"of":211,"metrics":{"PARAMS":"32.8M","Percentage correct":"89.46"},"uses_additional_data":true},{"leaderboard":"/sota/image-classification-on-cifar-100","task":"Image Classification","dataset":"CIFAR-100","model":"DenseNet-BC-190, S=4","rank_in_archive_order":45,"of":211,"metrics":{"PARAMS":"26.3M","Percentage correct":"87.44"},"uses_additional_data":true},{"leaderboard":"/sota/image-classification-on-cifar-100","task":"Image Classification","dataset":"CIFAR-100","model":"WRN-40-10, S=4","rank_in_archive_order":48,"of":211,"metrics":{"Percentage correct":"86.90"},"uses_additional_data":true},{"leaderboard":"/sota/image-classification-on-cifar-100","task":"Image Classification","dataset":"CIFAR-100","model":"WRN-28-10, S=4","rank_in_archive_order":59,"of":211,"metrics":{"Percentage correct":"85.74"},"uses_additional_data":true},{"leaderboard":"/sota/image-classification-on-imagenet","task":"Image Classification","dataset":"ImageNet","model":"SE-ResNeXt-101, 64x4d, S=2(320px)","rank_in_archive_order":417,"of":1060,"metrics":{"GFLOPs":"38.2","Number of params":"98M","Top 1 Accuracy":"83.6%"},"uses_additional_data":false},{"leaderboard":"/sota/image-classification-on-imagenet","task":"Image Classification","dataset":"ImageNet","model":"SE-ResNeXt-101, 64x4d, S=2(416px)","rank_in_archive_order":434,"of":1060,"metrics":{"GFLOPs":"61.1","Number of params":"98M","Top 1 Accuracy":"83.34%"},"uses_additional_data":false},{"leaderboard":"/sota/image-classification-on-imagenet","task":"Image Classification","dataset":"ImageNet","model":"ResNeXt-101, 64x4d, S=2(224px)","rank_in_archive_order":571,"of":1060,"metrics":{"GFLOPs":"18.8","Number of params":"88.6M","Top 1 Accuracy":"82.13%"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=2011.14660","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}