|MAIN :: RESEARCH :: PUBLICATIONS|
I completed my PhD in Electrical and Computer Engineering at Cornell University in August 2014, under the supervision of Prof. Salman Avestimehr and I am part of the Foundations of Information Engineering (FoIE) group. I received my B.S. in Mathematics and ECE from the Worcester Polytechnic Institute, in Worcester, MA, in 2009. I spent the summer of 2011 as an intern at HP Labs, in Palo Alto, CA.
Starting in Fall 2014, I will be a postdoctoral researcher working with Prof. Tom Courtade (UC Berkeley) and Prof. David Tse (Stanford).
I am mainly interested in problems in Information Theory, Communication Theory and Coding. I am also broadly interested in Probability Theory and Optimization. I've been focusing most of my research in Network Information Theory problems, particularly related to multi-user multi-hop wireless networks.
Simons Fellowship for Spring 2015 - Simons Institute for the Theory of Computing
Qualcomm Innovation Fellowship, 2013 (with Alireza Vahid)
Olin Fellowship - Cornell University, 2009-2010
Provost's MQP Award (Senior Thesis Award) - WPI Math Department, 2009
Journal Papers and Preprints
[J1] I. Shomorony and A. S. Avestimehr, "Two-Unicast Wireless Networks: Characterizing the Degrees of Freedom", IEEE Transactions on Information Theory, Vol 59, No 1, January 2013.
[J2] I. Shomorony and A. S. Avestimehr, "Worst-Case Additive Noise in Wireless Networks", IEEE Transactions on Information Theory, Vol 59, No 6, June 2013.
[J3] I. Shomorony, R. Etkin, F. Parvaresh and A. S. Avestimehr, "Diamond Networks with Bursty Traffic: Bounds on the Minimum Energy-Per-Bit", IEEE Transactions on Information Theory, Vol 60, No 1, January 2014.
[J4] H. Asnani, I. Shomorony, A. S. Avestimehr and T. Weissman, "Network Compression: Worst-Case Analysis", submitted to IEEE Transactions on Information Theory.
[J5] R. Etkin, F. Parvaresh, I. Shomorony and A. S. Avestimehr, "On Min-Cut Algorithms for Half-Duplex Relay Networks", submitted to IEEE Transactions on Information Theory.
[J6] I. Shomorony and A. S. Avestimehr, "Degrees of Freedom of Two-Hop Wireless Networks: Everyone Gets the Entire Cake", IEEE Transactions on Information Theory, Vol 60, No 5, May 2014.
[C1] I. Shomorony and A. S. Avestimehr, "Sum Degrees-of-Freedom of Two-Unicast Wireless Networks", ISIT 2011. (Student Paper Award Finalist)
[C2] I. Shomorony, R. Etkin, F. Parvaresh and A. S. Avestimehr, "Bounds on the Minimum Energy-Per-Bit for Bursty Traffic in Diamond Networks", ISIT 2012.
[C3] I. Shomorony and A. S. Avestimehr, "Is Gaussian Noise the Worst-Case Additive Noise in Wireless Networks?", ISIT 2012.
[C4] I. Shomorony, A. S. Avestimehr, H. Asnani and T. Weissman, "Worst-Case Source for Distributed Compression with Quadratic Distortion", ITW 2012.
[C5] I. Shomorony and A. S. Avestimehr, "On the Role of Deterministic Models in K x K x K Wireless Networks", ITW 2012.
[C6] I. Shomorony and A. S. Avestimehr, "Degrees of Freedom of Two-Hop Wireless Networks: Everybody Gets the Entire Cake", Allerton 2012. (Slides)
[C7] H. Asnani, I. Shomorony, A. S. Avestimehr and T. Weissman, "Network Compression: Worst-Case Analysis", ISIT 2013.
[C8] R. Etkin, F. Parvaresh, I. Shomorony and A. S. Avestimehr, "On Efficient Min-Cut Approximations in Half-Duplex Relay Networks", ISIT 2013.
[C9] I. Shomorony and A. S. Avestimehr, "A Generalized Cut-Set Bound for Deterministic Multi-Flow Networks and its Applications", ISIT 2014.
Fundamentals of Multi-hop Multi-flow Wireless Networks [J1],[C1],[J6],[C5],[C6]
Recent years have seen a dramatic increase in the wireless data traffic, caused by the success of online media streaming services and the proliferation of smart phones, tablets, and netbooks. Given the scarcity of wireless spectrum, the only way to meet this ever-increasing demand is to exploit a much denser spatial reuse of the spectrum by considering new wireless network architectures; in particular those based on multi-hop and multi-flow paradigms. However, little is known about the fundamental principles that govern the design of communication schemes for multi-hop multi-flow systems, and, in most of these scenarios, an exact characterization of the Shannon capacity is still out of the question. Thus, in this research project, we seek alternative ways to study these networks, such as (i) formulating and studying deterministic models that mimic the behavior of their stochastic counterparts, and (ii) considering the high-SNR capacity approximation provided by a degrees of freedom analysis.
The characterization of the degrees of freedom often leads to a conceptual understanding of fundamental aspects of communication in these networks. This is the case, for instance, of our results in [J6]. By showing that K degrees of freedom can be achieved on a two-hop K x K x K network, we provide an answer to a conceptual question about distributed MIMO systems which can be formulated in an algebraic way as a diagonalization problem, illustrated below.
If the K relays could cooperate (i.e., if they were a single MIMO node), they would apply the linear transformation in order to diagoanalize the end-to-end network transform. But if the K relays cannot cooperate, how can this end-to-end diagonalization be obtained in a distributed way?
Robustness of Theoretical Models [J2],[C3],[C4],[C5],[J4],[C7]
Gaussian models are ubiquitous in data compression and data communication problems. The additive noise experienced by wireless receivers, for instance, is often modeled as a white Gaussian random process. Similarly, but perhaps less intuitively, data sources are also commonly modeled as Gaussian processes. While these models are formally justified in point-to-point setups as the worst-case assumptions, the same was not known to be the case in network setups, and the main reason for these assumptions was analytical tractability. Thus, from a theoretical standpoint, a relevant question is: In what scenarios are these Gaussian models worst-case assumptions? And, from a practical perspective: Can compression and communication schemes be designed under Gaussian assumptions and still be useful in non-Gaussian scenarios?
We answered these questions in the context of data communication in wireless networks [J2] and joint source-channel coding in arbitrary networks [J5]. We proved that the Gaussian distribution is indeed worst-case in these cases, by providing a framework that allows coding schemes designed under Gaussian assumptions to be converted to coding schemes that are robust in the sense that they achieve the same performance under arbitrary statistical assumptions. The figure below illustrates how this is done in network compression problems [J5].
Each source node applies a transformation to its non-Gaussian data source with the purpose of "gaussifying" it. More precisely, we find a sequence of such transformations such that the resulting effective sources converge in distribution to Gaussian, i.e.,
All network nodes will then operate as if the sources were indeed Gaussian, and the destinations will apply the inverse transformations to the reconstructed sequences, to "ungaussify" them. We show that there exist optimal coding schemes for this network for which the above convergence in distribution implies convergence in distortion, i.e.,
Besides settling the aforementioned questions, this result and the result in [J2] allow us to establish connections between the distortion (or capacity) regions of networks under different models. In [C5], we pursued this direction and demonstrated that in two-hop multi-flow wireless networks the capacity under the Gaussian model can be upper bounded by the capacity of the network under a deterministic model.
Relay Networks with Real-World Constraints [J3],[C2], [J5],[C8]
The study of wireless systems is traditionally performed with simplified models whose goal is to capture the fundamental aspects of communication and provide insights into the design of optimal communication strategies. However, particularly for the case of large wireless relay networks, there are big discrepancies between these theoretical models and the practical systems, which makes the conversion from theory to practice a research effort in itself. Examples of these discrepancies include full duplex versus half duplex antennas, and the assumption of availability of channel state information at the network nodes.
The issues of synchronization between network nodes and energy-efficient communication were addressed in [J3] in the context of a two-relay network. The main motivation for this work are wireless sensor networks, where nodes operate on batteries and the communication of data tends to be bursty, i.e., intermittent. In this scenario, synchronization techniques must be used before every data transmission, and the synchronization energy costs become relevant. In [J3], by approximately characterizing the minimum energy-per-bit required in this asynchronous scenario, we were able to prove the near optimality of training sequences for synchronization, and determine, for a given two-node network, what is the optimal relay selection, as illustrated below.
In the figures above, if relay R2 were in a green area, the optimal relay selection (from an energy point of view) would be R1 and R2, if it were in a red area, the optimal relay selection would be only R1, and if it were in a blue area, the optimal relay selection would be only R2. The yellow regions correspond to points where our characterization is not tight enough to determine the optimal relay selection.
This research project comprises many of our ongoing and future research directions. In particular, we have been studying how results on degrees of freedom such as [J1] and [J6] are affected by real-world constraints such as computational complexity and limited channel diversity. For example, the recent work in [Issa] tackles these two issues by characterizing the degrees of freedom achievable on a 2 x 2 x 2 wireless network with linear schemes and no channel diversity. We are currently studying how these ideas can be scaled for the general K x K x K setting.