{"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/embedded-binarized-neural-networks","title":"Embedded Binarized Neural Networks","arxiv_id":"1709.02260","date":"2017-09-06","proceeding":null,"authors":["Bradley McDanel","Surat Teerapittayanon","H. T. Kung"],"abstract":"We study embedded Binarized Neural Networks (eBNNs) with the aim of allowing\ncurrent binarized neural networks (BNNs) in the literature to perform\nfeedforward inference efficiently on small embedded devices. We focus on\nminimizing the required memory footprint, given that these devices often have\nmemory as small as tens of kilobytes (KB). Beyond minimizing the memory\nrequired to store weights, as in a BNN, we show that it is essential to\nminimize the memory used for temporaries which hold intermediate results\nbetween layers in feedforward inference. To accomplish this, eBNN reorders the\ncomputation of inference while preserving the original BNN structure, and uses\njust a single floating-point temporary for the entire neural network. All\nintermediate results from a layer are stored as binary values, as opposed to\nfloating-points used in current BNN implementations, leading to a 32x reduction\nin required temporary space. We provide empirical evidence that our proposed\neBNN approach allows efficient inference (10s of ms) on devices with severely\nlimited memory (10s of KB). For example, eBNN achieves 95\\% accuracy on the\nMNIST dataset running on an Intel Curie with only 15 KB of usable memory with\nan inference runtime of under 50 ms per sample. To ease the development of\napplications in embedded contexts, we make our source code available that\nallows users to train and discover eBNN models for a learning task at hand,\nwhich fit within the memory constraint of the target device.","url_abs":"http://arxiv.org/abs/1709.02260v1","url_pdf":"http://arxiv.org/pdf/1709.02260v1.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":"embedded-binarized-neural-networks","repo_url":"https://gitlab.com/htkung/ddnn","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"none","reach":null},{"paper_slug":"embedded-binarized-neural-networks","repo_url":"https://github.com/DHLSan/PYNQeBNN","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":null}],"tasks":[],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}