Distributed Optimization Algorithms for Sensor Networks and Data Centers

There is a pressing need for creative solutions for dealing with the so-called Big Data: enormous quantities of high-dimensional data generated across several societal domains. One solution involves storing the data in different computers, because the data is either too large to fit in the memory of a single computer or is generated locally, as in a sensor network. The data is then processed in a distributed way, with computers exchanging estimates of the solution of a given task rather than database entries or private measurements. Although this distributed processing has the added benefit of preserving privacy of local databases, the main drawback is the increase in processing time, as communication links can be slow. Given a task, a distributed algorithm will thus be efficient if it uses a small number of communications.

In this project, the student will become familiar with the state-of-the-art distributed algorithms for optimization, with an emphasis on algorithms that reconstruct sparse signals, with applications in machine learning and image processing. A new algorithm will then be designed, implemented, and compared against the state-of-the-art. Familiarity with Matlab, Julia, or Python is desirable.


[1] J. F. C. Mota, J. M. F. Xavier, P. M. Q. Aguiar, M. PĆ¼schel, "Distributed Basis Pursuit", IEEE Transactions on Signal Processing, Vol. 60, No. 4, pp. 1942-1956, 2012 (http://dx.doi.org/10.1109/TSP.2011.2182347).

[2] W. Shi, Q. Ling, G. Wu, W. Yin, "EXTRA: An Exact First-Order Algorithm for Decentralized Consensus Optimization", SIAM J. Optim., Vol. 25, No. 2, pp. 944-966, 2015 (https://doi.org/10.1137/14096668X).

[3] S. Patterson, Y. C. Eldar, I. Keidar, "Distributed Compressed Sensing for Static and Time-Varying Networks", IEEE Transactions on Signal Processing, Vol. 62, No. 19, pp. 4931-4946, 2014 (https://doi.org/10.1109/TSP.2014.2340812).

Supervisor name: 
Dr. Joao Mota
Supervisor and Deputy email addresses: 

Project Type: