SPARC · Sparse Regression Codes
FP7 — People (Marie Curie Actions)
- Duration
- 2014-03-01 → 2018-02-28
- EU contribution
- €100,000
- Participants
- 1
- Scheme
- MC-CIG
Lines connect the coordinator with its partners.
Results in brief
Sparse Regression Codes
Project: Sparse Regression Codes PI: Ramji Venkataramanan, University Lecturer, University of Cambridge (rv285@cam.ac.uk) Project website: http://www2.eng.cam.ac.uk/~rv285/sparc.html The aim of the project was to develop powerful low-complexity codes for communication and compression. Communication schemes usually constructed by combining a modulation technique with a binary error-correcting code. Though simple to implement, the rates delivered by this approach is far from optimal. One of the main ideas of the project was to step back from the coding/modulation divide, and instead the statistical framework of sparse linear regression to design codes. In the project, we have designed codes for both communication and compression with near-optimal rates and low-complexity encoding and decoding algorithms. The research outputs of the project include new code designs, theoretical analysis of the proposed codes, and software implementation of the designed Sparse Regression Codes (SPARCs). Specific research highlights of the projects include: i) Design and rigorous analysis of a novel SPARC decoder based on `approximate message passing' (AMP), ii) New algorithms for power allocation to optimize finite length error performance of SPARCs, iii) Rigorous theoretical analysis of lossy compression using SPARCs, showing that SPARCs attain the optimal distortion vs rate trade-off for Gaussian sources, iv) Novel SPARC-inspired techniques for high-dimensional statistical estimation, such as new variants of James-Stein estimators. The CIG has helped the PI to grow his research group in Cambridge, and start new collaborations. In particular, the grant has funded one post-doc (Dr K. P. Srinath), part-funded one PhD student (A. Greig), and facilitated productive collaborations with Dr C. Rush ( Columbia University), Prof. A. Montanari (Stanford), and Prof. O. Johnson (Bristol). The work done in this project formed the basis for the PI's successful EPSRC First Grant proposal and a matching award from the Isaac Newton Trust. In summary, the CIG has been instrumental in jump-starting the PI's career in Cambridge, and the integration of the PI into the international research community in his field.
Data: CORDIS, © European Union
Project objective
Modern communication networks are constantly growing in size, speed and sophistication. Applications such as streaming multimedia and cloud computing consume ever-increasing amounts of bandwidth in wireless networks and the Internet. New kinds of networks are being built for sensing, communication and coordination in applications as diverse as transportation, security, power grids, and infrastructure monitoring. All these applications demand a high degree of reliability and energy efficiency, in addition to having low delay tolerance. To meet these demands in the face of rapidly growing data volume, it is critical to have fast, rate-optimal codes for data transmission and compression.The project aims to develop low-complexity, rate-optimal codes using the framework of high-dimensional sparse regression. Using the sparse regression methodology, we will construct codes whose rates approach the optimal information-theoretic limits with low-complexity coding algorithms for a large class of communication and compression problems. This class includes Gaussian channels and sources, which are important in practice. First, the codes will be designed for the basic problems of point-to-point communication and lossy compression. The codes will then serve as building blocks which can be combined to implement coding schemes for various network settings involving distributed communication and compression. The final goal, therefore, is to develop a library of low-complexity, rate-optimal codes for a variety of network models such as multi-access, broadcast, and interference channels.""
Original text from CORDIS.
Participants
- THE CHANCELLOR MASTERS AND SCHOLARS OF THE UNIVERSITY OF CAMBRIDGE · CAMBRIDGECoordinatorUnited Kingdom
Links
Data: CORDIS, © European Union