Collaborative Research: MLWiNS: Physical Layer Communication revisited via Deep Learning

合作研究：MLWiNS：通过深度学习重新审视物理层通信

• 批准号: 2002664
• 负责人: Sewoong Oh
• 金额: $ 22.33万
• 依托单位: University of Washington
• 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: 2020-08-01 至 2023-07-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2002664&HistoricalAwards=false
• 关键词: Collaborative; Research; MLWiNS; Physical; Layer

Reliable communication is a workhorse of the modern information age. The disciplines of communication, coding, and information theory drive the innovation by designing efficient codes that allow transmissions to be robustly decoded. Progress in near optimal codes is made by individual human ingenuity and breakthroughs have been, befittingly, sporadic, spread over several decades. Deep learning has recently shown strong promise in problems where the space of algorithmic choices is enormous (e.g., Go). This scenario likewise characterizes communication theory. Deep learning methods can play a crucial role in achieving the aforementioned goals. All resulting algorithms will be maintained on an online repository with full source code and documentation. The practical applications of the new codes will be explored in the context of wireless deployments. The research outcomes will be used to develop new undergraduate and graduate curricula. The fundamental nature of the research spans two areas of independent scientific and technical interest: finite block length information theory and the mathematics of deep learning. This project aims to bring the tools of deep learning to design a new family of encoding and decoding methods for canonical communication models; the codes so generated are naturally built for finite block lengths. In parallel, the viewpoint of neural network architectures as encoding and decoding procedures provides a unique vantage point to study their mathematical properties.This 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.

可靠的通信是现代信息时代的一项重要工作。通信、编码和信息理论的学科通过设计允许传输被鲁棒解码的有效代码来驱动创新。近最优编码的进展是由个人的聪明才智和突破已经取得了，恰当地，零星的，几十年来传播。深度学习最近在算法选择空间巨大的问题中表现出了强大的前景（例如，Go）。这种情况同样也是传播理论的特点。深度学习方法可以在实现上述目标方面发挥关键作用。所有产生的算法将保持在一个在线存储库的完整源代码和文档。将在无线部署的背景下探讨新代码的实际应用。研究成果将用于开发新的本科生和研究生课程。 该研究的基本性质跨越两个独立的科学和技术领域：有限块长度信息理论和深度学习数学。该项目旨在利用深度学习工具为规范通信模型设计一系列新的编码和解码方法;这样生成的代码自然是为有限块长度构建的。与此同时，神经网络架构作为编码和解码过程的观点为研究其数学特性提供了一个独特的Vantage位置。该奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

项目成果

CRISP: Curriculum based Sequential Neural Decoders for Polar Code Family

• DOI: 10.48550/arxiv.2210.00313
• 发表时间: 2022-10
• 期刊: ArXiv
• 作者: Ashwin Hebbar;Viraj Nadkarni;Ashok Vardhan Makkuva;S. Bhat;Sewoong Oh;P. Viswanath

Machine Learning-Aided Efficient Decoding of Reed–Muller Subcodes

机器学习辅助 Reed Muller 子码的高效解码

• DOI: 10.1109/jsait.2023.3298362
• 发表时间: 2023
• 期刊: IEEE Journal on Selected Areas in Information Theory
• 作者: Jamali, Mohammad Vahid;Liu, Xiyang;Makkuva, Ashok Vardhan;Mahdavifar, Hessam;Oh, Sewoong;Viswanath, Pramod
• 通讯作者: Viswanath, Pramod