{"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/branchynet-fast-inference-via-early-exiting","title":"BranchyNet: Fast Inference via Early Exiting from Deep Neural Networks","arxiv_id":"1709.01686","date":"2017-09-06","proceeding":null,"authors":["Surat Teerapittayanon","Bradley McDanel","H. T. Kung"],"abstract":"Deep neural networks are state of the art methods for many learning tasks due\nto their ability to extract increasingly better features at each network layer.\nHowever, the improved performance of additional layers in a deep network comes\nat the cost of added latency and energy usage in feedforward inference. As\nnetworks continue to get deeper and larger, these costs become more prohibitive\nfor real-time and energy-sensitive applications. To address this issue, we\npresent BranchyNet, a novel deep network architecture that is augmented with\nadditional side branch classifiers. The architecture allows prediction results\nfor a large portion of test samples to exit the network early via these\nbranches when samples can already be inferred with high confidence. BranchyNet\nexploits the observation that features learned at an early layer of a network\nmay often be sufficient for the classification of many data points. For more\ndifficult samples, which are expected less frequently, BranchyNet will use\nfurther or all network layers to provide the best likelihood of correct\nprediction. We study the BranchyNet architecture using several well-known\nnetworks (LeNet, AlexNet, ResNet) and datasets (MNIST, CIFAR10) and show that\nit can both improve accuracy and significantly reduce the inference time of the\nnetwork.","url_abs":"http://arxiv.org/abs/1709.01686v1","url_pdf":"http://arxiv.org/pdf/1709.01686v1.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":"branchynet-fast-inference-via-early-exiting","repo_url":"https://gitlab.com/htkung/branchynet","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"none","reach":null},{"paper_slug":"branchynet-fast-inference-via-early-exiting","repo_url":"https://github.com/gorakraj/earlyexit_onnx","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null},{"paper_slug":"branchynet-fast-inference-via-early-exiting","repo_url":"https://github.com/yangyucheng000/Paper-3/tree/main/One2Branch/MindSpore","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"mindspore","reach":null}],"tasks":[],"methods":[{"method_slug":"1x1-convolution","method_name":"1x1 Convolution"},{"method_slug":"convolution","method_name":"Convolution"},{"method_slug":"dense-connections","method_name":"Dense Connections"},{"method_slug":"dropout","method_name":"Dropout"},{"method_slug":"early-exiting","method_name":"Early exiting"},{"method_slug":"grouped-convolution","method_name":"Grouped Convolution"},{"method_slug":"local-response-normalization","method_name":"Local Response Normalization"},{"method_slug":"max-pooling","method_name":"Max Pooling"},{"method_slug":"relu","method_name":"ReLU"},{"method_slug":"softmax","method_name":"Softmax"}],"datasets_introduced":[],"methods_introduced":[{"slug":"early-exiting","name":"Early exiting","full_name":"Early exiting using confidence measures"}],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1709.01686","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}