Generalized Submodular Optimization

Team

Faculty: Yixin Chen, Chenyang Lu

PhD Student: Abusayeed Saifullah, Chengjie Wu

Alumni: Sangeeta Bhattacharya, You Xu

---

While wireless sensor networks (WSNs) have traditionally been used as a specialized platform for single applications, recent years have witnessed the emergence of integrated wireless sensor networks as shared infrastructure for multiple applications. The evolution of wireless sensor networks from dedicated platforms to shared infrastructure is driven by a wide range of integrated sensing systems such as urban sensing, building automation, and environmental monitoring. Compared to a separate WSN dedicated to each application, a shared WSN offers more flexibility, adaptivity, and cost- effectiveness through dynamic resource and node allocation.

Due to their resource constraints in bandwidth, memory, and energy, shared sensor networks face a critical need for optimizing the Quality of Monitoring (QoM) through dynamic resource allocation to the contending applications. These emerging QoM optimization problems in shared sensor networks are computationally challenging, as they are discrete and nonlinear. Furthermore, the optimization approach must: 1) deal with multiple resource constraints, 2) scale to large networks, and 3) handle network and environmental dynamics.

To address these challenges, it is important to exploit the special structure of these optimiza- tion problems for sensing applications. A key observation is that most of the QoM functions of the physical phenomena exhibit diminishing returns when more nodes are allocated to an application, a property known as submodularity. Submodular properties are ubiquitous in distributed sensing applications and have been addressed in the literature. However, most of the existing works that utilize submodularity rely on unique problem structures and restrictive assumptions only appli- cable to specific applications. Moreover, all existing submodular optimization algorithms assume a centralized control, which limits their scalability. To ensure scalability and deal with network dynamics, it is essential to develop distributed and online algorithms for submodular optimization. In this project, both centralized and distributed algorithms are proposed to solve submodular optimization problems in WSNs. We develop novel distributed approaches that exploit submodularity. By incorporating submodularity into a market-based framework, we provide the first approximation bounds for distributed submodular optimization.

Publications

• C. Wu, Y. Xu, Y. Chen and C. Lu, Submodular Game for Distributed Application Allocation in Shared Sensor Networks, IEEE International Conference on Computer Communications (INFOCOM'12), March 2012. [PDF]

• Y. Xu, A. Saifullah, Y. Chen, and C. Lu, Near Optimal Multi-Application Allocation in Shared Sensor Networks, The 11th ACM International Symposium on Mobile Ad Hoc Networking and Computing (MobiHoc 2010), Chicago, Illinois; Sept. 2010; pp. 181--190. [PDF]

• S. Bhattacharya, A. Saifullah, C. Lu, and G. C. Roman, Multi-Application Deployment in Shared Sensor Networks Based on Quality of Monitoring, The 17th IEEE Real-Time and Embedded Technology and Applications Symposium (RTAS 2010), Stockholm, Sweden; April 2010; pp. 259--268. [PDF]

If you have any questions or comments, feel free to email Chengjie Wu.

Acknowledgements

This work is supported by the NSF under NeTS Grant CNS-0627126.