{"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/blockqnn-efficient-block-wise-neural-network","title":"BlockQNN: Efficient Block-wise Neural Network Architecture Generation","arxiv_id":"1808.05584","date":"2018-08-16","proceeding":null,"authors":["Zhao Zhong","Zichen Yang","Boyang Deng","Junjie Yan","Wei Wu","Jing Shao","Cheng-Lin Liu"],"abstract":"Convolutional neural networks have gained a remarkable success in computer\nvision. However, most usable network architectures are hand-crafted and usually\nrequire expertise and elaborate design. In this paper, we provide a block-wise\nnetwork generation pipeline called BlockQNN which automatically builds\nhigh-performance networks using the Q-Learning paradigm with epsilon-greedy\nexploration strategy. The optimal network block is constructed by the learning\nagent which is trained to choose component layers sequentially. We stack the\nblock to construct the whole auto-generated network. To accelerate the\ngeneration process, we also propose a distributed asynchronous framework and an\nearly stop strategy. The block-wise generation brings unique advantages: (1) it\nyields state-of-the-art results in comparison to the hand-crafted networks on\nimage classification, particularly, the best network generated by BlockQNN\nachieves 2.35% top-1 error rate on CIFAR-10. (2) it offers tremendous reduction\nof the search space in designing networks, spending only 3 days with 32 GPUs. A\nfaster version can yield a comparable result with only 1 GPU in 20 hours. (3)\nit has strong generalizability in that the network built on CIFAR also performs\nwell on the larger-scale dataset. The best network achieves very competitive\naccuracy of 82.0% top-1 and 96.0% top-5 on ImageNet.","url_abs":"http://arxiv.org/abs/1808.05584v1","url_pdf":"http://arxiv.org/pdf/1808.05584v1.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":"blockqnn-efficient-block-wise-neural-network","repo_url":"https://github.com/gomerudo/nas-dmrl","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok","spdx":"MIT"}},{"paper_slug":"blockqnn-efficient-block-wise-neural-network","repo_url":"https://github.com/gomerudo/nas-env","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"unanswered"}}],"tasks":[{"task_slug":null,"task_name":"GPU"},{"task_slug":"image-classification","task_name":"Image Classification"},{"task_slug":"q-learning","task_name":"Q-Learning"},{"task_slug":"image-classification","task_name":"image-classification"}],"methods":[{"method_slug":"q-learning","method_name":"Q-Learning"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1808.05584","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}