[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-120015-en":3,"doc-seo-120015-105":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":4,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":27,"seo_description":14,"update_tm":28,"read_time":29},120015,137441390410,"Hazel","https://ap-avatar.wpscdn.com/avatar/2000252f4ab5702993?_k=1776741390130283984",6,"Technology","NASH - Neural Architecture Search for Hardware-Optimized Machine Learning Models","Machine learning models deployed in real applications must balance high accuracy with high throughput and low latency, often under tight hardware and power constraints. This paper proposes NASH, a hardware-aware neural architecture search method that targets superior accuracy while maintaining low hardware utilization and efficient inference. Four NASH variants guide training by selecting hardware-friendly operations within convolutional neural networks. Experiments on ImageNet using ResNet18/34 show up to 3.1% top-1 and 2.2% top-5 accuracy gains over non-NASH baselines, and improved accuracy–hardware trade-offs. The approach is automated through integration with the FINN hardware model synthesis tool, and performance reaches up to 324.5 fps with open-source code.","NASH: Neural Architecture Search for Hardware-Optimized  \nMachine Learning Models  \nMengfei Ji, Yuchun Chang, Baolin Zhang and Zaid Al-Ars  \nJilin University  \nChangchun, Jilin, China  \nDelft University of Technology  \nDelft, The Netherlands  \nEmail: [jimengfei007@outlook.com](jimengfei007@outlook.com)  \narXiv :2403 .01845v2 [ cs .LG] 10 Mar 2024  \nAbstract—As machine learning (ML) algorithms get deployed in anever-increasing number of applications, these algorithms need to achieve better trade-offs between high accuracy, high throughput and low latency. This paper introduces NASH, a novel approach that applies neural architecture search to machine learning hardware. Using NASH, hardware designs can achieve not only high throughput and low latency but also superior accuracy performance. We present four versions of the NASH strategy in this paper, all of which show higher accuracy than the original models. The strategy can be applied to various convolutional neural networks, selecting specific model operations among many to guide the training process toward higher accuracy. Experimental results show that applying NASH on ResNet18 or ResNet34 achieves a top 1 accuracy increase of up to 3.1% and a top 5 accuracy increase of up to 2.2% compared to the non-NASH version when tested on the ImageNet data set. We also integrated this approach into the FINN hardware model synthesis tool to automate the application of our approach and the generation of the hardware model. Results show that using FINN can achieve a maximum throughput of 324.5 fps. In addition, NASH models can also result in a better trade-off between accuracy and hardware resource utilization. The accuracy-hardware (HW) Pareto curve shows that the models with the four NASH versions represent the best trade-offs achieving the highest accuracy for a given HW utilization. The code for our implementation is open-source and publicly available on GitHub at [https://github.com/MFJI/NASH](https://github.com/MFJI/NASH).  \nIndex Terms—CNN, NAS, FINN, HW, ResNet  \nI. INTRODUCTION  \nThe rise of ML has impacted many aspects of our daily lives from web searches to autonomous driving [1][2][3] . In many situations, the application imposes strict demands on ML model behavior, requiring high accuracy, high throughput, low latency, low HW utilization, and/or low power. Designing custom hardware implementations of these models can help achieve many of these requirements. However, hardware implementations require the use of low-bit width data, which leads to a decrease in the accuracy of the model, thereby limiting the applicability of such solutions in practice. To ensure the wide applicability of custom hardware ML model implementations, we need to investigate advanced methods to improve the accuracy of low-bit-width models (e.g., binarized models) .  \nExisting solutions focus on software-based improvement of these models using multi-weight binary models [4], or linear combinations of binary weights [5], or improving gradient paths[6] . Another approach is to use neural architecture search (NAS) to increase the accuracy of the networks [7] . Although these methods are able to improve model accuracy, they are fully software-based and cannot be used directly to improve the accuracy of hardware implementations. In this paper, we propose NASH, a hardware-based architecture search algorithm that optimizes the architecture of an ML model specifically for hardware implementations. The design uses the opensource hardware compiler FINN [8] [9], which is an efficient framework to explore deep neural network inference on FPGAs specifically  \nfor quantized neural networks. The implementation ensures high accuracy and low HW utilization of the generated model for the used hardware. Four versions of the NASH strategy are shown in this paper and are able to achieve the best trade-off between accuracy and HW resource utilization.  \nThe contributions of this paper are as follows:  \n1) We propose an approac","cbCaig9N2Ehyp7PB","https://ap.wps.com/l/cbCaig9N2Ehyp7PB","pdf",1431908,1,9,"English","en",105,"# Introduction\n## Background\n## Neural Architecture Search\n## FINN Framework\n# NASH Method\n## Hardware-Aware Architecture Search\n# Experimental Results\n## Accuracy vs Hardware Trade-off\n# Conclusion","[{\"question\":\"What problem does NASH address for deployed machine learning models?\",\"answer\":\"NASH addresses the need to achieve better trade-offs between high accuracy, high throughput, low latency, and low hardware utilization when ML models are implemented on hardware.\"},{\"question\":\"How does NASH improve model accuracy for hardware implementations?\",\"answer\":\"NASH applies neural architecture search specifically for hardware, selecting model operations that steer training toward higher accuracy while keeping hardware efficiency.\"},{\"question\":\"What tools and models were used to validate NASH?\",\"answer\":\"The paper integrates NASH into the FINN hardware model synthesis tool and evaluates it on hardware-optimized ResNet18 and ResNet34 using ImageNet benchmarks.\"}]","NASH - Neural Architecture Search for Hardware-Optimized Machine Learning Models | PDF",1785727738,23,{"code":4,"msg":31,"data":32},"ok",{"site_id":24,"language":23,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":28},"nash-neural-architecture-search-for-hardware-optimized-machine-learning-models","",{"@graph":36,"@context":85},[37,54,68],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,48,51],{"item":41,"name":42,"@type":43,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":47},"https://docshare.wps.com/document/","Document",2,{"item":49,"name":12,"@type":43,"position":50},"https://docshare.wps.com/document/technology/",3,{"item":52,"name":13,"@type":43,"position":53},"https://docshare.wps.com/document/nash-neural-architecture-search-for-hardware-optimized-machine-learning-models/120015/",4,{"url":52,"name":13,"@type":55,"author":56,"headline":13,"publisher":58,"fileFormat":61,"inLanguage":23,"description":14,"dateModified":62,"datePublished":62,"encodingFormat":61,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":57},"Person",{"url":41,"name":59,"@type":60},"DocShare","Organization","application/pdf","2026-08-03",true,{"@type":65,"interactionType":66,"userInteractionCount":4},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"What problem does NASH address for deployed machine learning models?","Question",{"text":75,"@type":76},"NASH addresses the need to achieve better trade-offs between high accuracy, high throughput, low latency, and low hardware utilization when ML models are implemented on hardware.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does NASH improve model accuracy for hardware implementations?",{"text":80,"@type":76},"NASH applies neural architecture search specifically for hardware, selecting model operations that steer training toward higher accuracy while keeping hardware efficiency.",{"name":82,"@type":73,"acceptedAnswer":83},"What tools and models were used to validate NASH?",{"text":84,"@type":76},"The paper integrates NASH into the FINN hardware model synthesis tool and evaluates it on hardware-optimized ResNet18 and ResNet34 using ImageNet benchmarks.","https://schema.org",{"og:url":52,"og:type":87,"og:title":13,"og:site_name":59,"og:description":14},"article",{"robots":89,"canonical":52},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":92},[93,97,101,105,110,113,118,123,127,130,134],{"id":20,"doc_module":4,"doc_module_name":46,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":47,"doc_module":4,"doc_module_name":46,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":53,"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":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":111,"slug":112},50,"technology",{"id":114,"doc_module":4,"doc_module_name":46,"category_name":115,"show_sort_weight":116,"slug":117},7,"Healthcare",40,"healthcare",{"id":119,"doc_module":4,"doc_module_name":46,"category_name":120,"show_sort_weight":121,"slug":122},8,"Research & Report",30,"research-report",{"id":21,"doc_module":4,"doc_module_name":46,"category_name":124,"show_sort_weight":125,"slug":126},"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"]