{"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/understanding-and-improving-convolutional","title":"Understanding and Improving Convolutional Neural Networks via Concatenated Rectified Linear Units","arxiv_id":"1603.05201","date":"2016-03-16","proceeding":null,"authors":["Wenling Shang","Kihyuk Sohn","Diogo Almeida","Honglak Lee"],"abstract":"Recently, convolutional neural networks (CNNs) have been used as a powerful\ntool to solve many problems of machine learning and computer vision. In this\npaper, we aim to provide insight on the property of convolutional neural\nnetworks, as well as a generic method to improve the performance of many CNN\narchitectures. Specifically, we first examine existing CNN models and observe\nan intriguing property that the filters in the lower layers form pairs (i.e.,\nfilters with opposite phase). Inspired by our observation, we propose a novel,\nsimple yet effective activation scheme called concatenated ReLU (CRelu) and\ntheoretically analyze its reconstruction property in CNNs. We integrate CRelu\ninto several state-of-the-art CNN architectures and demonstrate improvement in\ntheir recognition performance on CIFAR-10/100 and ImageNet datasets with fewer\ntrainable parameters. Our results suggest that better understanding of the\nproperties of CNNs can lead to significant performance improvement with a\nsimple modification.","url_abs":"http://arxiv.org/abs/1603.05201v2","url_pdf":"http://arxiv.org/pdf/1603.05201v2.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":"understanding-and-improving-convolutional","repo_url":"https://github.com/fengwang/CReLU","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok","spdx":"BSD-3-Clause"}},{"paper_slug":"understanding-and-improving-convolutional","repo_url":"https://github.com/Andy-Dang12/Activation-Functions","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}}],"tasks":[],"methods":[{"method_slug":"adam","method_name":"Adam"},{"method_slug":"crelu","method_name":"CReLU"},{"method_slug":"convolution","method_name":"Convolution"},{"method_slug":"dense-connections","method_name":"Dense Connections"},{"method_slug":"dropout","method_name":"Dropout"},{"method_slug":"max-pooling","method_name":"Max Pooling"},{"method_slug":"relu","method_name":"ReLU"},{"method_slug":"sgd-with-momentum","method_name":"SGD with Momentum"},{"method_slug":"softmax","method_name":"Softmax"},{"method_slug":"weight-decay","method_name":"Weight Decay"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1603.05201","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}