CNS Core: Small: Bringing the Cloud to the Masses: Models, Algorithms, and Systems Targeting Deep Learning for the IoT Era

CNS 核心：小：将云带入大众：针对物联网时代深度学习的模型、算法和系统

• 批准号: 2007284
• 负责人: Radu Marculescu
• 金额: $ 49.9万
• 依托单位: University of Texas at Austin
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2020
• 资助国家: 美国
• 起止时间: 2020-10-01 至 2024-09-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2007284&HistoricalAwards=false
• 关键词: CNS; Core; Small; Bringing; Cloud

An estimated one trillion Internet-of-Things (IoT) devices are expected to impact multiple market segments by 2035. Such a rapid growth in IoT devices necessitates new breakthroughs in machine learning approaches to fully exploit the compute power offered by these devices. However, the significant computational requirements of existing deep learning approaches represent a major bottleneck for their large-scale deployment on hardware-constrained edge devices. To improve this state of affairs, this project seeks to develop new models, algorithms, and prototypes that can enable faster and computationally inexpensive inference on resource-constrained mobile/IoT devices. These new models can ultimately help preserve user’s privacy by conducting inference on a network of edge devices instead of sending data to the cloud.Through its proposed research and curriculum, this project enables a complete symbiosis between software, architectures, and system meant to transform not just parallel computing, but also real life applications ranging from data centers, to IoT and personal devices. Indeed, developing such “lightweight” machine learning systems can ultimately enable the democratization of access to application-specific hardware and make such systems widely available. All these contributions aim at helping transitioning traditional cloud computing into computing at the edge. Through curriculum related innovations, student competitions, and other outreach activities, this project is poised to raise awareness and increase participation of various underrepresented groups to this area of engineeringThis award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

Please try later.

项目成果

Efficient On-Device Training via Gradient Filtering

• DOI: 10.1109/cvpr52729.2023.00371
• 发表时间: 2023-01
• 期刊: 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)
• 作者: Yuedong Yang-;Guihong Li;R. Marculescu

Medical Image Segmentation via Cascaded Attention Decoding

• DOI: 10.1109/wacv56688.2023.00616
• 发表时间: 2023-01
• 期刊: 2023 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)
• 作者: Md Mostafijur Rahman;R. Marculescu

Multi-scale Hierarchical Vision Transformer with Cascaded Attention Decoding for Medical Image Segmentation

• DOI: 10.48550/arxiv.2303.16892
• 发表时间: 2023-03
• 期刊: ArXiv
• 作者: Md Mostafijur Rahman;R. Marculescu

A Hardware Prototype Targeting Distributed Deep Learning for On-device Inference

针对设备端推理的分布式深度学习的硬件原型

• DOI: 10.1109/cvprw50498.2020.00207
• 发表时间: 2020
• 期刊: Computer Vision and Pattern Recognition
• 作者: Farcas, Allen-Jasmin;Li, Guihong;Bhardwaj, Kartikeya;Marculescu, Radu
• 通讯作者: Marculescu, Radu

ZiCo: Zero-shot NAS via Inverse Coefficient of Variation on Gradients

• DOI: 10.48550/arxiv.2301.11300
• 发表时间: 2023-01
• 期刊: ArXiv
• 作者: Guihong Li;Yuedong Yang-;Kartikeya Bhardwaj;R. Marculescu