{"id":19543172,"url":"https://github.com/yeonghyeon/shake-shake","last_synced_at":"2026-05-20T10:04:28.969Z","repository":{"id":159176006,"uuid":"232964897","full_name":"YeongHyeon/Shake-Shake","owner":"YeongHyeon","description":"TensorFlow implementation of Shake-Shake Regularization.","archived":false,"fork":false,"pushed_at":"2020-02-04T00:18:43.000Z","size":1206,"stargazers_count":2,"open_issues_count":0,"forks_count":0,"subscribers_count":3,"default_branch":"master","last_synced_at":"2025-02-26T05:32:55.584Z","etag":null,"topics":["cnn","convolutional-neural-networks","regularization","tensorflow"],"latest_commit_sha":null,"homepage":"","language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/YeongHyeon.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null}},"created_at":"2020-01-10T04:23:21.000Z","updated_at":"2024-06-27T00:34:39.000Z","dependencies_parsed_at":"2023-07-28T19:00:38.595Z","dependency_job_id":null,"html_url":"https://github.com/YeongHyeon/Shake-Shake","commit_stats":null,"previous_names":[],"tags_count":0,"template":false,"template_full_name":null,"purl":"pkg:github/YeongHyeon/Shake-Shake","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/YeongHyeon%2FShake-Shake","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/YeongHyeon%2FShake-Shake/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/YeongHyeon%2FShake-Shake/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/YeongHyeon%2FShake-Shake/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/YeongHyeon","download_url":"https://codeload.github.com/YeongHyeon/Shake-Shake/tar.gz/refs/heads/master","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/YeongHyeon%2FShake-Shake/sbom","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":263849780,"owners_count":23519719,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2022-07-04T15:15:14.044Z","host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["cnn","convolutional-neural-networks","regularization","tensorflow"],"created_at":"2024-11-11T03:17:48.213Z","updated_at":"2026-05-20T10:04:28.906Z","avatar_url":"https://github.com/YeongHyeon.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"Shake-Shake Regularization\n=====\n\nTensorFlow implementation of Shake-Shake Regularization.  \n\n## Concept\n\u003cdiv align=\"center\"\u003e\n  \u003cimg src=\"./figures/shake.png\" width=\"600\"\u003e  \n  \u003cp\u003eThe concept of Shake-Shake Regularization [1].\u003c/p\u003e\n\u003c/div\u003e\n\n## Procedure\n\nThe whole procedure for using Shake-Shake Regularization is shown as below. All the figures are redesigned by \u003ca href=\"https://github.com/YeongHyeon\"\u003eYeongHyeon\u003c/a\u003e.  \n\n\u003cdiv align=\"center\"\u003e\n  \u003cimg src=\"./figures/phase0.png\" width=\"600\"\u003e  \n  \u003cp\u003ePhase 0. Preparing for Shake-Shake.\u003c/p\u003e\n\u003c/div\u003e\n\n\u003cdiv align=\"center\"\u003e\n  \u003cimg src=\"./figures/phase1.png\" width=\"600\"\u003e  \n  \u003cp\u003ePhase 1. Forward propagation in training.\u003c/p\u003e\n\u003c/div\u003e\n\n\u003cdiv align=\"center\"\u003e\n  \u003cimg src=\"./figures/phase2.png\" width=\"600\"\u003e  \n  \u003cp\u003ePhase 2. Backward propagation in training.\u003c/p\u003e\n\u003c/div\u003e\n\n\u003cdiv align=\"center\"\u003e\n  \u003cimg src=\"./figures/phase3.png\" width=\"600\"\u003e  \n  \u003cp\u003ePhase 3. Forward propagation in test.\u003c/p\u003e\n\u003c/div\u003e\n\n## Performance\n\nThe performance is measured using below two CNN architectures.\n\n\u003cdiv align=\"center\"\u003e\n  \u003cimg src=\"./figures/cnn.png\" width=\"600\"\u003e  \n  \u003cp\u003eTwo Convolutional Neural Networks for experiment.\u003c/p\u003e\n\u003c/div\u003e\n\n| |ConvNet8|ConvNet8 with S-S|\n|:---|:---:|:---:|\n|Accuracy|0.99340|0.99420|\n|Precision|0.99339|0.99416|\n|Recall|0.99329|0.99413|\n|F1-Score|0.99334|0.99414|\n\n### ConvNet8\n```\nConfusion Matrix\n[[ 979    0    0    0    0    0    0    1    0    0]\n [   0 1132    0    1    0    0    1    1    0    0]\n [   0    0 1029    0    0    0    0    3    0    0]\n [   0    0    1 1006    0    3    0    0    0    0]\n [   0    0    1    0  975    0    2    0    0    4]\n [   1    0    0    7    0  882    1    0    0    1]\n [   4    2    0    0    0    1  950    0    1    0]\n [   1    3    3    2    0    0    0 1018    1    0]\n [   3    0    1    1    0    1    0    0  966    2]\n [   0    0    0    1    6    2    0    3    0  997]]\nClass-0 | Precision: 0.99089, Recall: 0.99898, F1-Score: 0.99492\nClass-1 | Precision: 0.99560, Recall: 0.99736, F1-Score: 0.99648\nClass-2 | Precision: 0.99420, Recall: 0.99709, F1-Score: 0.99565\nClass-3 | Precision: 0.98821, Recall: 0.99604, F1-Score: 0.99211\nClass-4 | Precision: 0.99388, Recall: 0.99287, F1-Score: 0.99338\nClass-5 | Precision: 0.99213, Recall: 0.98879, F1-Score: 0.99045\nClass-6 | Precision: 0.99581, Recall: 0.99165, F1-Score: 0.99372\nClass-7 | Precision: 0.99220, Recall: 0.99027, F1-Score: 0.99124\nClass-8 | Precision: 0.99793, Recall: 0.99179, F1-Score: 0.99485\nClass-9 | Precision: 0.99303, Recall: 0.98811, F1-Score: 0.99056\n\nTotal | Accuracy: 0.99340, Precision: 0.99339, Recall: 0.99329, F1-Score: 0.99334\n```\n\n### ConvNet8 with S-S (ConvNet8 + Shake-Shake Regularization)\n```\nConfusion Matrix\n[[ 978    1    0    0    0    0    0    1    0    0]\n [   0 1131    0    0    0    0    2    1    1    0]\n [   1    1 1027    0    0    0    0    2    1    0]\n [   0    0    0 1008    0    2    0    0    0    0]\n [   0    0    0    0  979    0    1    0    0    2]\n [   1    0    0    6    0  884    1    0    0    0]\n [   3    2    0    0    2    1  948    0    2    0]\n [   0    1    4    0    1    0    0 1020    1    1]\n [   2    0    2    0    0    1    0    0  967    2]\n [   0    0    0    0    4    3    0    1    1 1000]]\nClass-0 | Precision: 0.99289, Recall: 0.99796, F1-Score: 0.99542\nClass-1 | Precision: 0.99560, Recall: 0.99648, F1-Score: 0.99604\nClass-2 | Precision: 0.99419, Recall: 0.99516, F1-Score: 0.99467\nClass-3 | Precision: 0.99408, Recall: 0.99802, F1-Score: 0.99605\nClass-4 | Precision: 0.99290, Recall: 0.99695, F1-Score: 0.99492\nClass-5 | Precision: 0.99214, Recall: 0.99103, F1-Score: 0.99159\nClass-6 | Precision: 0.99580, Recall: 0.98956, F1-Score: 0.99267\nClass-7 | Precision: 0.99512, Recall: 0.99222, F1-Score: 0.99367\nClass-8 | Precision: 0.99383, Recall: 0.99281, F1-Score: 0.99332\nClass-9 | Precision: 0.99502, Recall: 0.99108, F1-Score: 0.99305\n\nTotal | Accuracy: 0.99420, Precision: 0.99416, Recall: 0.99413, F1-Score: 0.99414\n```\n\n## Requirements\n* Python 3.6.8  \n* Tensorflow 1.14.0  \n* Numpy 1.17.1  \n* Matplotlib 3.1.1  \n\n## Reference\n[1] Gastaldi, Xavier. \u003ca href=\"https://arxiv.org/abs/1705.07485\"\u003eShake-Shake Regularization.\u003c/a\u003e arXiv preprint arXiv:1705.07485 (2017).\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fyeonghyeon%2Fshake-shake","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fyeonghyeon%2Fshake-shake","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fyeonghyeon%2Fshake-shake/lists"}