{"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/streamlined-deployment-for-quantized-neural","title":"Streamlined Deployment for Quantized Neural Networks","arxiv_id":"1709.04060","date":"2017-09-12","proceeding":null,"authors":["Yaman Umuroglu","Magnus Jahre"],"abstract":"Running Deep Neural Network (DNN) models on devices with limited\ncomputational capability is a challenge due to large compute and memory\nrequirements. Quantized Neural Networks (QNNs) have emerged as a potential\nsolution to this problem, promising to offer most of the DNN accuracy benefits\nwith much lower computational cost. However, harvesting these benefits on\nexisting mobile CPUs is a challenge since operations on highly quantized\ndatatypes are not natively supported in most instruction set architectures\n(ISAs). In this work, we first describe a streamlining flow to convert all QNN\ninference operations to integer ones. Afterwards, we provide techniques based\non processing one bit position at a time (bit-serial) to show how QNNs can be\nefficiently deployed using common bitwise operations. We demonstrate the\npotential of QNNs on mobile CPUs with microbenchmarks and on a quantized\nAlexNet, which is 3.5x faster than an optimized 8-bit baseline. Our bit-serial\nmatrix multiplication library is available on GitHub at https://git.io/vhshn","url_abs":"http://arxiv.org/abs/1709.04060v2","url_pdf":"http://arxiv.org/pdf/1709.04060v2.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":"streamlined-deployment-for-quantized-neural","repo_url":"https://github.com/EECS-NTNU/bismo","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":{"status":"ok","spdx":"BSD-3-Clause"}}],"tasks":[],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1709.04060","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}