[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-86240-en":3,"doc-seo-86240-105":30,"detail-sidebar-cat-0-en-105":91},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":21,"is_downloadable":21,"audit_status":21,"page_count":22,"language":23,"language_code":24,"site_id":25,"html_lang":24,"table_of_contents":26,"faqs":27,"seo_title":13,"seo_description":14,"update_tm":28,"read_time":29},86240,1374391974585,"Genevieve","https://ap-avatar.wpscdn.com/davatar_276721f389ce27ea32af1340a28f341c",8,"Research & Report","Towards Efficient Convolutional Neural Network for Embedded Hardware via Multi-Dimensional Pruning","该文提出面向嵌入式硬件的多维度CNN剪枝方法，通过在深度、宽度与分辨率等三个维度上识别冗余单元，实现更高的模型效率与精度平衡。方法以两步评估框架为核心：先用内维度评估策略（INES）快速确定各维度最冗余候选，从而大幅降低后续代价；再用跨维度评估策略（ITES）将复杂度、准确率与推理延迟整合为统一指标，做出全局冗余判断。进一步设计启发式剪枝算法迭代移除冗余单元，在巨大的搜索空间中寻找适合资源受限边缘设备的小型架构，并在ImageNet与嵌入式推理场景获得更优的精度与速度表现。","Towards Efficient Convolutional Neural Network for Embedded Hardware via Multi-Dimensional Pruning  \nHao Kong 1,2 , Di Liu3 , Xiangzhong Luo 1 , Shuo Huai 1,2 , Ravi Subramaniam4 ,  \nChristian Makaya4 , Qian Lin4 , and Weichen Liu 1*  \n1 School of Computer Science and Engineering, Nanyang Technological University, Singapore  \n2HP-NTU Digital Manufacturing Corporate Lab, Nanyang Technological University, Singapore 3Department of Computer Science, Norwegian University of Science and Technology, Trondheim, Norway.  \n4HP Inc., Palo Alto, California, USA  \nJul  \nI. INTRODUCTION Over the past decade, the evolution of deep convolutional neural  \nbeen benefiting many deep learning applications a growing demand to deploy advanced CNNs at  \nFig. 1: Comparison of different methods in terms of model MACs, accuracy, and inference latency. The baseline network is ResNet50 [1], which is trained on ImageNet. The latency is measured on AGX Xavier with a power budget of 30W.  \naccurately identify redundant units in the three dimensions, we first propose an inter-dimensional evaluation strategy (ITES) to comprehensively evaluate the importance of pruning units across different dimensions. In ITES, we integrate the contribution of each pruning unit to model complexity, accuracy, and inference latency into a unified metric, global importance, according to which the unit with the lowest global importance is considered redundant and can be safely removed. However, as ITES needs to collect multiple metrics for comprehensive evaluation, directly applying ITES to traverse all units of the three dimensions will be extremely time-consuming. To this end, we also introduce an inner-dimensional evaluation strategy (INES) to first quickly evaluate units within each dimension and identify the most redundant unit of each dimension. By this means, ITES only needs tobe performed on the most redundant unit of each dimension for the final pruning decision. INES reduces the pruning candidates for each dimension from multiple to one, which reduces the evaluation cost of ITES and enhances the pruning efficiency significantly. On top of the two-step evaluation framework composed by INES and ITES, we design a heuristic pruning algorithm, which utilizes INES and ITESto progressively identify and prune redundant units. In this way, we can efficiently search for the optimal tiny architecture for resourceconstrained embedded devices in the huge design space formed by the three dimensions.  \nOur main contributions are three-fold:  \n1) We introduce an inter-dimensional importance evaluation strategy (ITES) to evaluate the importance of units across different dimensions. We integrate the contribution of each unit to model complexity, accuracy and latency into a comprehensive metric,  \nFig. 2: The overview of TECO, where INES evaluates the local importance of units inside each dimension and ITES evaluates the global importance of units across different dimensions. The inner-dimensional evaluation is skipped for the resolution dimension (See Section III-B for the detailed reason) .  \nglobal importance, which enables us to accurately identify redundant units in the three dimensions.  \n2) We also propose an inner-dimensional importance evaluation strategy (INES) to quickly evaluate the local importance of units within each dimension, which reduces the pruning choices for each dimension from multiple to one, alleviating the evaluation overhead of ITES and improving the pruning efficiency of our framework.  \n3) Based on ITES and INES, we design a heuristic pruning algorithm to progressively prune the three dimensions of CNNs. By iteratively identifying and removing redundant units with INES and ITES, our pruning algorithm can efficiently find the optimal tiny model for edge devices in the huge design space of multi-dimensional pruning.  \nAs shown in Fig. 1, our TECO obtains 3.97% higher top-1 accuracy on ImageNet than HRank [8] with similar MACs. For on-device acceleration, TECO is","cbCaiv7DtotAGIdD","https://ap.wps.com/l/cbCaiv7DtotAGIdD","pdf",748325,2,1,7,"English","en",105,"# Introduction\n# Related Work\n# Multi-Dimensional","[{\"question\":\"本文面向嵌入式硬件的多维度剪枝目标是什么？\",\"answer\":\"目标是在深度、宽度与分辨率等多个维度上移除冗余单元，以在压缩率提升的同时尽量保持精度，并降低推理延迟与模型复杂度。\"},{\"question\":\"INES与ITES在剪枝框架中分别承担什么作用？\",\"answer\":\"INES用于在每个维度内快速评估并找出最冗余的候选单元，减少候选数量与评估开销；ITES用于跨维度综合复杂度、准确率与延迟形成统一的全局重要性指标，指导最终冗余判断。\"},{\"question\":\"为什么使用两步评估（先INES后ITES）能提升剪枝效率？\",\"answer\":\"因为ITES需要收集多个指标做全局评估，若对三维所有单元逐个评估会非常耗时。INES先把每个维度的候选从多个压缩到一个最冗余单元，使得ITES只需在关键候选上执行，从而显著降低时间成本并提升效率。\"}]",1784209734,18,{"code":4,"msg":31,"data":32},"ok",{"site_id":25,"language":24,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":28},"towards-efficient-convolutional-neural-network-for-embedded-hardware-via-multi-dimensional-pruning","",{"@graph":36,"@context":85},[37,53,68],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,47,50],{"item":41,"name":42,"@type":43,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":20},"https://docshare.wps.com/document/","Document",{"item":48,"name":12,"@type":43,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":43,"position":52},"https://docshare.wps.com/document/towards-efficient-convolutional-neural-network-for-embedded-hardware-via-multi-dimensional-pruning/86240/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":24,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":41,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-27","2026-07-16",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"本文面向嵌入式硬件的多维度剪枝目标是什么？","Question",{"text":75,"@type":76},"目标是在深度、宽度与分辨率等多个维度上移除冗余单元，以在压缩率提升的同时尽量保持精度，并降低推理延迟与模型复杂度。","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"INES与ITES在剪枝框架中分别承担什么作用？",{"text":80,"@type":76},"INES用于在每个维度内快速评估并找出最冗余的候选单元，减少候选数量与评估开销；ITES用于跨维度综合复杂度、准确率与延迟形成统一的全局重要性指标，指导最终冗余判断。",{"name":82,"@type":73,"acceptedAnswer":83},"为什么使用两步评估（先INES后ITES）能提升剪枝效率？",{"text":84,"@type":76},"因为ITES需要收集多个指标做全局评估，若对三维所有单元逐个评估会非常耗时。INES先把每个维度的候选从多个压缩到一个最冗余单元，使得ITES只需在关键候选上执行，从而显著降低时间成本并提升效率。","https://schema.org",{"og:url":51,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":51},"index,follow",{"doc_id":7,"site_id":25},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,119,122,127,130,134],{"id":21,"doc_module":4,"doc_module_name":46,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":20,"doc_module":4,"doc_module_name":46,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":46,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":46,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":46,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":22,"doc_module":4,"doc_module_name":46,"category_name":116,"show_sort_weight":117,"slug":118},"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":120,"slug":121},30,"research-report",{"id":123,"doc_module":4,"doc_module_name":46,"category_name":124,"show_sort_weight":125,"slug":126},9,"Religion & Spirituality",20,"religion-spirituality",{"id":125,"doc_module":4,"doc_module_name":46,"category_name":128,"show_sort_weight":125,"slug":129},"World Cup","world-cup",{"id":131,"doc_module":4,"doc_module_name":46,"category_name":132,"show_sort_weight":131,"slug":133},10,"Lifestyle","lifestyle",{"id":135,"doc_module":4,"doc_module_name":46,"category_name":136,"show_sort_weight":106,"slug":137},19,"General","general"]