SIST Professor Published in IEEE Communications

ON2018-08-31TAG: ShanghaiTech UniversityCATEGORY: School of Information Science and Technology

Professor Shi YuanMing and his collaborators at HKUST and Tsinghua University have recently developed large-scale sparse and low-rank frameworks for optimizing across the communication, computation and storage resources in ultra-dense networks (UDNs), thereby providing principled ways to design communication-efficient mobile artificial intelligence (AI) systems and intelligent internet-of-things (IoT) networks. Their exciting results were recently published as “Generalized Sparse and Low-Rank Optimization for Ultra-Dense Networks” by Shi et al. in IEEE Communications.

Structured Sparse Optimization for Large-Scale Network Adaptation

In UDNs, large-scale network adaptation plays a pivotal role in effectively utilizing densely deployed radio access points to support massive mobile devices. For various network adaptation problems in UDNs, the solution vector is expected to be sparse in a structured manner, e.g., radio access point selection results in a group sparsity structure. To illustrate the powerfulness of the generalized sparse representation and scalable optimization paradigms, the researchers presented representative examples of group sparse beamforming for green Cloud-RANs, and structured sparse optimization for active user detection and user admission control. In particular, their work on green Cloud-RANs won the 2016 IEEE Marconi Prize Paper Award.

Generalized Low-Rank Optimization with Network Side Information

UDNs are highly complex to optimize, and it is critical to exploit the available network side information. For examples, network connectivity information, cached content at the access points, and locally computed intermediate values for wireless distributed computing, all serve as exploitable side information for efficiently designing coding and decoding in UDNs. The researchers provided a generalized low-rank matrix modeling framework to exploit the network side information, which helped to efficiently optimize across the communication, computation, and storage resources. A general low-rank optimization problem was further established by incorporating the network side information to design communication efficient UDNs.

Optimization Algorithms and Analysis

To address the algorithmic challenges for the sparse and low-rank modeling frameworks for UDNs, for the general convex optimization problems, the researchers developed a principled two-stage framework with the capability of providing certificates of infeasibility, enabling parallel and scalable computing via the operator splitting method. For the nonconvex low-rank optimization problems, they developed a Riemannian optimization framework by exploiting the manifold geometry of fixed-rank matrices to achieve fast convergence rates and scalability. Their work on large-scale convex optimization paper won the 2016 IEEE Signal Processing Society Young Author Best Paper Award.

Their work was supported by grants from the National Natural Science Foundation of China, the Shanghai Municipal Science and Technology Commission, and ShanghaiTech University.

Read more at:

Figure 1. Group sparse beamforming for green Cloud-RAN design.

Figure 2. Generalized low-rank model: (a) topological interference management; (b) cache-aided interference channel.

Figure 3. A schematic view of Riemannian optimization framework.