{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2025,12,21]],"date-time":"2025-12-21T07:11:44Z","timestamp":1766301104176,"version":"3.41.0"},"reference-count":29,"publisher":"Association for Computing Machinery (ACM)","issue":"2","license":[{"start":{"date-parts":[[2022,3,10]],"date-time":"2022-03-10T00:00:00Z","timestamp":1646870400000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/www.acm.org\/publications\/policies\/copyright_policy#Background"}],"content-domain":{"domain":["dl.acm.org"],"crossmark-restriction":true},"short-container-title":["J. Emerg. Technol. Comput. Syst."],"published-print":{"date-parts":[[2022,4,30]]},"abstract":"<jats:p>\n            While gliomas have become the most common cancerous brain tumors, manual diagnoses from 3D MRIs are time-consuming and possibly inconsistent when conducted by different radiotherapists, which leads to the pressing demand for automatic segmentation of brain tumors. State-of-the-art approaches employ FCNs to automatically segment the MRI scans. In particular, 3D U-Net has achieved notable performance and motivated a series of subsequent works. However, their significant size and heavy computation have impeded their actual deployment. Although there exists a body of literature on the compression of CNNs using low-precision representations, they either focus on storage reduction without computational improvement or cause severe performance degradation. In this article, we propose a CNN training algorithm that approximates weights and activations using\n            <jats:italic>non-negative integers<\/jats:italic>\n            along with\n            <jats:italic>trained affine mapping<\/jats:italic>\n            functions. Moreover, our approach allows the dot-product operations to be performed in an\n            <jats:italic>integer-arithmetic<\/jats:italic>\n            manner and defers the floating-point decoding and encoding phases until the end of layers. Experimental results on BraTS 2018 show that our trained affine mapping approach achieves near full-precision dice accuracy with 8-bit weights and activations. In addition, we achieve a dice accuracy within 0.005 and 0.01 of the full-precision counterparts when using 4-bit and 2-bit precisions, respectively.\n          <\/jats:p>","DOI":"10.1145\/3495210","type":"journal-article","created":{"date-parts":[[2022,3,10]],"date-time":"2022-03-10T14:06:32Z","timestamp":1646921192000},"page":"1-16","update-policy":"https:\/\/doi.org\/10.1145\/crossmark-policy","source":"Crossref","is-referenced-by-count":4,"title":["Optimizing 3D U-Net-based Brain Tumor Segmentation with Integer-arithmetic Deep Learning Accelerators"],"prefix":"10.1145","volume":"18","author":[{"given":"Weijia","family":"Wang","sequence":"first","affiliation":[{"name":"University of California, San Diego, La Jolla, CA"}]},{"given":"Bill","family":"Lin","sequence":"additional","affiliation":[{"name":"University of California, San Diego, La Jolla, CA"}]}],"member":"320","published-online":{"date-parts":[[2022,3,10]]},"reference":[{"key":"e_1_3_2_2_2","article-title":"Scalable methods for 8-bit training of neural networks","volume":"1805","author":"Banner Ron","year":"2018","unstructured":"Ron Banner, Itay Hubara, Elad Hoffer, and Daniel Soudry. 2018. Scalable methods for 8-bit training of neural networks. CoRR abs\/1805.11046 (2018).","journal-title":"CoRR"},{"key":"e_1_3_2_3_2","doi-asserted-by":"publisher","DOI":"10.5555\/525960"},{"key":"e_1_3_2_4_2","article-title":"Deep learning with low precision by half-wave Gaussian quantization","volume":"1702","author":"Cai Zhaowei","year":"2017","unstructured":"Zhaowei Cai, Xiaodong He, Jian Sun, and Nuno Vasconcelos. 2017. Deep learning with low precision by half-wave Gaussian quantization. CoRR abs\/1702.00953 (2017).","journal-title":"CoRR"},{"key":"e_1_3_2_5_2","article-title":"BinaryNet: Training deep neural networks with weights and activations constrained to +1 or  \\( - \\) 1","volume":"1602","author":"Courbariaux Matthieu","year":"2016","unstructured":"Matthieu Courbariaux and Yoshua Bengio. 2016. BinaryNet: Training deep neural networks with weights and activations constrained to +1 or \\( - \\) 1. CoRR abs\/1602.02830 (2016).","journal-title":"CoRR"},{"key":"e_1_3_2_6_2","first-page":"315","volume-title":"Proceedings of the 14th International Conference on Artificial Intelligence and Statistics (Proceedings of Machine Learning Research)","volume":"15","author":"Glorot Xavier","year":"2011","unstructured":"Xavier Glorot, Antoine Bordes, and Yoshua Bengio. 2011. Deep sparse rectifier neural networks. In Proceedings of the 14th International Conference on Artificial Intelligence and Statistics (Proceedings of Machine Learning Research), Geoffrey Gordon, David Dunson, and Miroslav Dud\u00edk (Eds.), Vol. 15. PMLR, Fort Lauderdale, FL, 315\u2013323. Retrieved from http:\/\/proceedings.mlr.press\/v15\/glorot11a.html."},{"key":"e_1_3_2_7_2","article-title":"Compressing deep convolutional networks using vector quantization","volume":"1412","author":"Gong Yunchao","year":"2014","unstructured":"Yunchao Gong, Liu Liu, Ming Yang, and Lubomir D. Bourdev. 2014. Compressing deep convolutional networks using vector quantization. CoRR abs\/1412.6115 (2014).","journal-title":"CoRR"},{"key":"e_1_3_2_8_2","article-title":"Deep compression: Compressing deep neural network with pruning, trained quantization and Huffman coding","volume":"1510","author":"Han Song","year":"2015","unstructured":"Song Han, Huizi Mao, and William J. Dally. 2015. Deep compression: Compressing deep neural network with pruning, trained quantization and Huffman coding. CoRR abs\/1510.00149 (2015).","journal-title":"CoRR"},{"key":"e_1_3_2_9_2","unstructured":"Sergey Ioffe and Christian Szegedy. 2015. Batch normalization: Accelerating deep network training by reducing internal covariate shift. 448\u2013456. Retrieved from http:\/\/jmlr.org\/proceedings\/papers\/v37\/ioffe15.pdf."},{"key":"e_1_3_2_10_2","article-title":"nnU-Net: Self-adapting framework for U-Net-based medical image segmentation","volume":"1809","author":"Isensee Fabian","year":"2018","unstructured":"Fabian Isensee, Jens Petersen, Andr\u00e9 Klein, David Zimmerer, Paul F. Jaeger, Simon Kohl, Jakob Wasserthal, Gregor K\u00f6hler, Tobias Norajitra, Sebastian J. Wirkert, and Klaus H. Maier-Hein. 2018. nnU-Net: Self-adapting framework for U-Net-based medical image segmentation. CoRR abs\/1809.10486 (2018).","journal-title":"CoRR"},{"key":"e_1_3_2_11_2","article-title":"Quantization and training of neural networks for efficient integer-arithmetic-only inference","volume":"1712","author":"Jacob Benoit","year":"2017","unstructured":"Benoit Jacob, Skirmantas Kligys, Bo Chen, Menglong Zhu, Matthew Tang, Andrew G. Howard, Hartwig Adam, and Dmitry Kalenichenko. 2017. Quantization and training of neural networks for efficient integer-arithmetic-only inference. CoRR abs\/1712.05877 (2017).","journal-title":"CoRR"},{"key":"e_1_3_2_12_2","article-title":"Ternary weight networks","volume":"1605","author":"Li Fengfu","year":"2016","unstructured":"Fengfu Li and Bin Liu. 2016. Ternary weight networks. CoRR abs\/1605.04711 (2016).","journal-title":"CoRR"},{"key":"e_1_3_2_13_2","article-title":"Overcoming challenges in fixed point training of deep convolutional networks","volume":"1607","author":"Lin Darryl Dexu","year":"2016","unstructured":"Darryl Dexu Lin and Sachin S. Talathi. 2016. Overcoming challenges in fixed point training of deep convolutional networks. CoRR abs\/1607.02241 (2016).","journal-title":"CoRR"},{"key":"e_1_3_2_14_2","doi-asserted-by":"publisher","DOI":"10.1109\/TIT.1982.1056489"},{"key":"e_1_3_2_15_2","article-title":"Fully convolutional networks for semantic segmentation","volume":"1411","author":"Long Jonathan","year":"2014","unstructured":"Jonathan Long, Evan Shelhamer, and Trevor Darrell. 2014. Fully convolutional networks for semantic segmentation. CoRR abs\/1411.4038 (2014).","journal-title":"CoRR"},{"key":"e_1_3_2_16_2","doi-asserted-by":"publisher","DOI":"10.1109\/TMI.2014.2377694"},{"key":"e_1_3_2_17_2","doi-asserted-by":"crossref","unstructured":"Fausto Milletari Nassir Navab and Seyed-Ahmad Ahmadi. 2016. V-Net: Fully convolutional neural networks for volumetric medical image segmentation. (2016). arXiv:cs.CV\/1606.04797.","DOI":"10.1109\/3DV.2016.79"},{"key":"e_1_3_2_18_2","doi-asserted-by":"crossref","unstructured":"Andriy Myronenko. 2018. 3D MRI brain tumor segmentation using autoencoder regularization. (2018). arXiv:cs.CV\/1810.11654.","DOI":"10.1007\/978-3-030-11726-9_28"},{"key":"e_1_3_2_19_2","doi-asserted-by":"publisher","DOI":"10.1109\/IITSI.2010.74"},{"key":"e_1_3_2_20_2","unstructured":"Nvidia. 2018. NVIDIA Deep Learning Accelerator. Retrieved from http:\/\/nvdla.org\/."},{"key":"e_1_3_2_21_2","article-title":"Fixed-point representation & fractional math","volume":"9","author":"Oberstar Erick L.","year":"2007","unstructured":"Erick L. Oberstar. 2007. Fixed-point representation & fractional math. Oberstar Consult. 9 (2007).","journal-title":"Oberstar Consult."},{"key":"e_1_3_2_22_2","article-title":"Understanding the exploding gradient problem","volume":"1211","author":"Pascanu Razvan","year":"2012","unstructured":"Razvan Pascanu, Tomas Mikolov, and Yoshua Bengio. 2012. Understanding the exploding gradient problem. CoRR abs\/1211.5063 (2012).","journal-title":"CoRR"},{"key":"e_1_3_2_23_2","article-title":"3DQ: Compact quantized neural networks for volumetric whole brain segmentation","volume":"1904","author":"Paschali Magdalini","year":"2019","unstructured":"Magdalini Paschali, Stefano Gasperini, Abhijit Guha Roy, Michael Y.-S. Fang, and Nassir Navab. 2019. 3DQ: Compact quantized neural networks for volumetric whole brain segmentation. CoRR abs\/1904.03110 (2019).","journal-title":"CoRR"},{"key":"e_1_3_2_24_2","doi-asserted-by":"crossref","unstructured":"Olaf Ronneberger Philipp Fischer and Thomas Brox. 2015. U-Net: Convolutional networks for biomedical image segmentation. (2015). arXiv:cs.CV\/1505.04597.","DOI":"10.1007\/978-3-319-24574-4_28"},{"key":"e_1_3_2_25_2","doi-asserted-by":"publisher","DOI":"10.1109\/TMI.2010.2046908"},{"key":"e_1_3_2_26_2","article-title":"Training high-performance and large-scale deep neural networks with full 8-bit integers","volume":"1909","author":"Yang Yukuan","year":"2019","unstructured":"Yukuan Yang, Shuang Wu, Lei Deng, Tianyi Yan, Yuan Xie, and Guoqi Li. 2019. Training high-performance and large-scale deep neural networks with full 8-bit integers. CoRR abs\/1909.02384 (2019).","journal-title":"CoRR"},{"key":"e_1_3_2_27_2","doi-asserted-by":"publisher","DOI":"10.1007\/978-3-319-67389-9_32"},{"key":"e_1_3_2_28_2","article-title":"DoReFa-Net: Training low bitwidth convolutional neural networks with low bitwidth gradients","volume":"1606","author":"Zhou Shuchang","year":"2016","unstructured":"Shuchang Zhou, Zekun Ni, Xinyu Zhou, He Wen, Yuxin Wu, and Yuheng Zou. 2016. DoReFa-Net: Training low bitwidth convolutional neural networks with low bitwidth gradients. CoRR abs\/1606.06160 (2016).","journal-title":"CoRR"},{"key":"e_1_3_2_29_2","article-title":"Trained ternary quantization","volume":"1612","author":"Zhu Chenzhuo","year":"2016","unstructured":"Chenzhuo Zhu, Song Han, Huizi Mao, and William J. Dally. 2016. Trained ternary quantization. CoRR abs\/1612.01064 (2016).","journal-title":"CoRR"},{"key":"e_1_3_2_30_2","doi-asserted-by":"crossref","unstructured":"\u00d6zg\u00fcn \u00c7i\u00e7ek Ahmed Abdulkadir Soeren S. Lienkamp Thomas Brox and Olaf Ronneberger. 2016. 3D U-Net: Learning dense volumetric segmentation from sparse annotation. (2016). arXiv:cs.CV\/1606.06650.","DOI":"10.1007\/978-3-319-46723-8_49"}],"container-title":["ACM Journal on Emerging Technologies in Computing Systems"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/dl.acm.org\/doi\/10.1145\/3495210","content-type":"unspecified","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/dl.acm.org\/doi\/pdf\/10.1145\/3495210","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2025,6,17]],"date-time":"2025-06-17T20:12:02Z","timestamp":1750191122000},"score":1,"resource":{"primary":{"URL":"https:\/\/dl.acm.org\/doi\/10.1145\/3495210"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2022,3,10]]},"references-count":29,"journal-issue":{"issue":"2","published-print":{"date-parts":[[2022,4,30]]}},"alternative-id":["10.1145\/3495210"],"URL":"https:\/\/doi.org\/10.1145\/3495210","relation":{},"ISSN":["1550-4832","1550-4840"],"issn-type":[{"type":"print","value":"1550-4832"},{"type":"electronic","value":"1550-4840"}],"subject":[],"published":{"date-parts":[[2022,3,10]]},"assertion":[{"value":"2020-12-01","order":0,"name":"received","label":"Received","group":{"name":"publication_history","label":"Publication History"}},{"value":"2021-11-01","order":1,"name":"accepted","label":"Accepted","group":{"name":"publication_history","label":"Publication History"}},{"value":"2022-03-10","order":2,"name":"published","label":"Published","group":{"name":"publication_history","label":"Publication History"}}]}}