CIF: Small: Collaborative Research: Error Correction with Natural Redundancy

第1部分：项目的非技术性描述本项目研究通过使用数据的内部结构来消除数据中的错误的基本问题。它表明，当前数据存储系统中存储的海量数据具有非常丰富的结构；因此，通过充分利用它们进行纠错，可以显著提高数据存储系统的可靠性。该项目研究了该技术的几个基本方面，包括如何发现和表征各种类型数据的高度复杂的结构，如何使用它们来有效地纠正数据中的错误以提高数据存储系统的可靠性，如何将该技术与现有的基于增加数据外部冗余的纠错技术相结合，以及如何在实际的数据存储系统中实施该技术。该项目解决了现代社会的一个关键问题：如何确保数据能够大规模和长期可靠地存储。这项新技术有可能大幅提高信息基础设施的可靠性，信息基础设施经常访问大量数据，用于科学和工业计算。该项目本质上是跨学科的：它结合了包括信息论、机器学习、大数据分析和算法设计在内的多个科学领域，旨在教育学生并为下一代存储系统的劳动力发展做出贡献。该项目结合了严格的理论分析和重大的实际应用，以促进学术界和产业界之间的合作，并通过共同努力创造新的科学进步。第2部分：项目的技术描述本项目研究如何利用大数据中固有的冗余进行纠错。大数据的例子包括语言、图像、数据库等。固有的冗余与纠错码(ECC)集成在一起，以实现有效的纠错。其目标是将存储系统中的数据可靠性提升到下一个级别。为了实现这一目标，将开发新的技术来发现压缩和未压缩数据中的各种类型的固有冗余。将探索将固有冗余解码器和ECC解码器相结合以实现有效纠错的新方法。这个项目结合了纠错和机器学习，本质上是跨学科的。它将以多种方式扩展目前关于纠错的知识。首先，它使用自然语言处理和深度学习技术来发现大数据中适合纠错的新型冗余，这些冗余超出了现有的信源-信道联合编码知识。这包括已由各种压缩算法压缩的数据的冗余发现技术。其次，研究了ECC纠错码的译码算法，不仅考虑了规则ECC引入的冗余度，也考虑了非规则的固有冗余度。它扩展了现有的纠错方案，以在容量和计算复杂性方面将固有冗余的基本限制用于纠错。第三，将理论研究与实际系统相结合，为下一代存储和传输大数据的系统奠定基础。现代社会越来越依赖数字数据。随着每天产生的爆炸性数据量，必须在纠错方面取得进展，以赶上数据爆炸的速度。该项目旨在将数据可靠性显著提高到新的水平，在这个方向上的改进可以为现代社会的日常工作和生活带来极大的好处。这个项目是编码理论和机器学习之间的交叉学科，可以促进信息理论和计算机科学界之间的合作。该项目将严谨的理论分析与重大的实际应用相结合，以促进学术界和工业界之间的合作，并通过共同努力创造新的科学进步。拟议的研究将与工程教育相结合，为研究生和本科生开发新的课程，并让未被充分代表的国内和国际学生参与高级研究。将在国家/国际会议和期刊上积极宣传这一成果。

Part 1: Nontechnical description of the projectThis project studies the fundamental problem of removing errors from data by using internal structures of data. It shows that the vast amount of data stored in current data-storage systems possess very rich structures; therefore, by fully exploiting them for error correction, the reliability of data-storage systems can be improved significantly. The project studies several fundamental aspects of this technology, including how to discover and characterize the highly complex structures of various types of data, how to use them to correct errors in data efficiently to improve the reliability of data-storage systems, how to combine the technology with existing error-correction techniques that are based on adding external redundancy to data, and how to implement the technology in practical data-storage systems. This project addresses a critical issue of the modern society: how to ensure that data can be stored reliably at large scale and over a long time. The new technology has the potential to substantially improve the dependability of information infrastructure, which accesses vast amounts of data frequently for scientific and industrial computing. The project is interdisciplinary in nature: it combines multiple scientific fields including information theory, machine learning, big data analysis and algorithm design, and aims to educate students and contribute to workforce development for next-generation storage systems. The project conjugates rigorous theoretical analysis and significant practical applications, to foster collaboration between academia and industry, and create new scientific advances with combined efforts.Part 2: Technical description of the projectThis project studies how to use the inherent redundancy in big data for error correction. Examples of big data include languages, images, databases, and others. The inherent redundancy is integrated with error-correcting codes (ECC) for effective error correction. The objective is to elevate data reliability in storage systems to the next level. To achieve this goal, new techniques will be developed to discover various types of inherent redundancy in both compressed and uncompressed data. New approaches will be explored to combine inherent-redundancy decoders and ECC decoders for effective error correction. Fundamental limits of both capacity and computational complexity will be studied for error correction using inherent redundancy.This project combines error correction with machine learning and is interdisciplinary in nature. It will expand the current knowledge on error correction in multiple ways. First, it uses techniques in natural language processing and deep learning to discover new types of redundancy in big data that are suitable for error correction, and which extend beyond current knowledge in joint source-channel coding. This includes redundancy discovery techniques for data already compressed by various compression algorithms. Second, it explores decoding algorithms for ECCs with not only regular ECC-imposed redundancy, but also irregular inherent redundancy. It extends existing error correction schemes to cast the fundamental limits of inherent redundancy for error correction, in terms of both capacity and computational complexity. Third, by integrating a theoretical study with practical systems, a foundation can be laid for next-generation systems that store and transmit big data.Modern society relies increasingly heavily on digital data. With the explosive amount of data generated each day, it is essential to make advances in error correction that can catch the speed of data explosion. This project aims at improving data reliability significantly to the next level, and improvements in this direction can be highly beneficial to the daily work and life of the modern society. This project, being interdisciplinary between coding theory and machine learning, can foster collaboration between the information theory and computer science communities. The project combines rigorous theoretical analysis with significant practical applications, to foster collaboration between academia and industry, and create new scientific advances with combined efforts. The proposed research will be integrated with engineering education by developing new courses for graduate and undergraduate students, and involving under-represented, domestic and international students in advanced research. The results will be actively publicized in national/international conferences and journals.