{"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/efficient-architecture-search-by-network","title":"Efficient Architecture Search by Network Transformation","arxiv_id":"1707.04873","date":"2017-07-16","proceeding":null,"authors":["Han Cai","Tianyao Chen","Wei-Nan Zhang","Yong Yu","Jun Wang"],"abstract":"Techniques for automatically designing deep neural network architectures such\nas reinforcement learning based approaches have recently shown promising\nresults. However, their success is based on vast computational resources (e.g.\nhundreds of GPUs), making them difficult to be widely used. A noticeable\nlimitation is that they still design and train each network from scratch during\nthe exploration of the architecture space, which is highly inefficient. In this\npaper, we propose a new framework toward efficient architecture search by\nexploring the architecture space based on the current network and reusing its\nweights. We employ a reinforcement learning agent as the meta-controller, whose\naction is to grow the network depth or layer width with function-preserving\ntransformations. As such, the previously validated networks can be reused for\nfurther exploration, thus saves a large amount of computational cost. We apply\nour method to explore the architecture space of the plain convolutional neural\nnetworks (no skip-connections, branching etc.) on image benchmark datasets\n(CIFAR-10, SVHN) with restricted computational resources (5 GPUs). Our method\ncan design highly competitive networks that outperform existing networks using\nthe same design scheme. On CIFAR-10, our model without skip-connections\nachieves 4.23\\% test error rate, exceeding a vast majority of modern\narchitectures and approaching DenseNet. Furthermore, by applying our method to\nexplore the DenseNet architecture space, we are able to achieve more accurate\nnetworks with fewer parameters.","url_abs":"http://arxiv.org/abs/1707.04873v2","url_pdf":"http://arxiv.org/pdf/1707.04873v2.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":"efficient-architecture-search-by-network","repo_url":"https://github.com/han-cai/EAS","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok","spdx":"MIT"}},{"paper_slug":"efficient-architecture-search-by-network","repo_url":"https://github.com/han-cai/RL4AS_NetTrans","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok","spdx":"MIT"}},{"paper_slug":"efficient-architecture-search-by-network","repo_url":"https://github.com/han-cai/PathLevel-EAS","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}}],"tasks":[{"task_slug":"image-classification","task_name":"Image Classification"},{"task_slug":"architecture-search","task_name":"Neural Architecture Search"},{"task_slug":"reinforcement-learning","task_name":"Reinforcement Learning"},{"task_slug":"reinforcement-learning-1","task_name":"Reinforcement Learning (RL)"},{"task_slug":"reinforcement-learning-2","task_name":"reinforcement-learning"}],"methods":[{"method_slug":"1x1-convolution","method_name":"1x1 Convolution"},{"method_slug":"average-pooling","method_name":"Average Pooling"},{"method_slug":"batch-normalization","method_name":"Batch Normalization"},{"method_slug":"concatenated-skip-connection","method_name":"Concatenated Skip Connection"},{"method_slug":"convolution","method_name":"Convolution"},{"method_slug":"dense-block","method_name":"Dense Block"},{"method_slug":"dense-connections","method_name":"Dense Connections"},{"method_slug":"dropout","method_name":"Dropout"},{"method_slug":"global-average-pooling","method_name":"Global Average Pooling"},{"method_slug":"kaiming-initialization","method_name":"Kaiming Initialization"},{"method_slug":"max-pooling","method_name":"Max Pooling"},{"method_slug":"relu","method_name":"ReLU"},{"method_slug":"softmax","method_name":"Softmax"}],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/image-classification-on-cifar-10","task":"Image Classification","dataset":"CIFAR-10","model":"RL+NT","rank_in_archive_order":154,"of":265,"metrics":{"Percentage correct":"94.6"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1707.04873","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}