June Zhang



  • Carnegie Mellon University
    Ph.D. Electrical and Computer Engineering, 2015.
    Advisor: José M.F. Moura
  • Stanford University
    M.S. Electrical Engineering, 2008. NSF Graduate Fellow.
  • Georgia Institute of Technology
    B.S. Electrical and Computer Engineering, 2005. Summa Cum Laude.


  • Center for Disease Control
    Postdoctoral researcher (current)
  • Cargnegie Mellon University
    Doctoral student (Aug 2010-Oct 2015)
  • Tsinghua University
    Visiting scholar (Jan 2009-Dec 2009)
    Researched upper-body human pose detection from still photo using graph-cut skin detection algorithm.
  • San Jose Panasonic Research Laboratory
    Part-time design consultant (Apr 2007-Sept 2007)
    Prototyped multi-screen, touch-based interactive device
  • D2M
    Hardware intern.
  • Apple Computer
    Embedded input engineering intern.
  • NASA Goddard Space Center
    Undergraduate intern.


  • network science
  • graph theory
  • stasticial signal processing
  • stochastic process
  • network visualization
  • discrete optimization
  • machine learing
  • image processing
  • HCI (tangible user interface)
  • data analysis


  1. Los Alamos National Lab Science of Signatures participant
  2. Microsoft Azure Research Fellowship
  3. Three Minute Thesis Competition CMU Semifinalist
  4. National Science Foundation (NSF) Graduate Research Fellowship
  5. Georgia Hope Scholarship
  6. IEEE Atlanta Scholarship
  7. Boeing Merit Scholarship


  1. “Impact of Topology on Epidemics and Cascading Failures”, Sandia National Laboratory, Albuquerque, USA, June 8, 2015.
  2. “Impact of Topology on Dynamical Processes on Networks: Connectivity and Competition Matters,” Seminar Series, Santa Fe Institute, Santa Fe, USA, Apr 27, 2015.
  3. “Impact of Topology on Dynamical Processes on Networks,” IS&T Seminar Speakers Series, Los Alamos National Laboratory, Los Alamos, USA, Apr 22, 2015.
  4. “Who are the Most Vulnerable Agents in a Network?” Energy and Information Systems Seminar, Georgia Tech Research Institute, Atlanta, USA, Dec 5, 2014.
  5. “Graph Structure and Vulnerable Agents in a Network,” Business Technology Seminar, Tepper School of Business Carnegie Mellon University, Pittsburgh, USA, Nov 14, 2014.
  6. “Who are the Most Vulnerable Agents in a Network?” Energy and Information Systems Seminar, Carnegie Mellon University, Pittsburgh, USA, Oct 27, 2014.
  7. “Who are the Most Vulnerable Agents in a Network?” Machine Learning and the Social Sciences Seminar, Carnegie Mellon University, Pittsburgh, USA, Sept 29, 2014.
  8. “Topology and Network Diffusion Processes,” Electrical and Computer Eng., Instituto Superior Técnico, Lisbon, Portugal, May 20, 2014.

Koalas are adorable! Unfortunately, they are also a very vulnerable species. They can only eat eucalyptus leaves, which are toxic to all other animals. Even for koalas, they can only eat a few species of eucalyptus. The danger to koalas is habitat destruction. Their natural home ranges are the costal areas of Australia, precisely the part of Australia that humans also want to live. Koalas require large territories but more and more of these area are opening to suburbia development. So modern day koalas have to live with the very real dangers of car accidents and dog attacks not to mention the ever decreasing number of eatable eucalyptus trees.

I don’t have a lot of money to donate to the Koala Hospital in Port Macquarie, NSW, Australia and I can’t donate my time. But as a graduate student, I am friend with many other graduate students who speak 2nd, 3rd, or even 4th language. So I thought that I’d design a poster for animal conservation and ask my friends to translate the text into other languages. What I thought would be a simple copy and pasting project turned out to be a lot more difficult than I anticipated: Illustrator CS3 can not handle Farsi because it goes from right to left. I could not get transcription of Bengali to work in anything other than my browser. But the end result looks so cool! It is really fascinating how languages share similarity by geography.

Translations thanks to: Rohan Chabukswar, Sérgio Pequito, Milos Cvetkovic, Stefanos Baros, Jhi-Young Joo, Nikos Arechiga, Javad Mohammadi, Evgeny Toropov, Subhro Das, Joya Deri, Yuko Ishii, Nipun Popli.

































Network Processes

Interactions lead to complex behaviors. Infections goes from host to host resulting in epidemics. Blackouts are caused by cascading failures in the power grid. The spread of rumors, viralization of internet memes and videos, and traffic congestions are due to contagion-like phenomena where the behavior (or state) of an agent in the system affects the behaviors (or states) of neigbhoring agents in the system.

These phenomena are complex not only because they involve potentially hundreds, thousands, or even millions of individual agents. They are complex due to the dependencies between all the agents, whereas traditional analysis usually assumes that agents behave independently of one another. Network science models the structure of dependencies as weighted or unweighted, directed or undirected graphs.

My research is to understand and quantify how the topology of the network affects the behavior of these contagion-like phenomena. Intuitively, it is more difficult for contagion to spread in systems with sparsely connected communities than systems with densely connected communities. However, this is also affected by the rate at which contagions spread and the rate at which agents recover. The inclusion of networks and dyanmical process make the problem difficult to solve both mathematically and computationally. Many graph related problems are combinatorial problems; many of these problems are NP-hard.

Network process models account for both the network topology and the dynamical process. We developed the scaled SIS (susceptible-infected-susceptible) process, which models network processes with arbitrary, finite-size, unweighted, undirected networks. The scaled SIS process has a closed-form equilibrium distribution. Further, many inference questions related to the scaled SIS process can be solved in polynomial-time.

Who are the Most Susceptible?

In network processes, an agent state is affected by the states of its neighbors, which are affected by the states of their neighbors; an agent's topological characteristics and its rates determines its susceptiblity to infection.

Using the scaled SIS process, we can determine for a given network topology, the set of agents (blue) that are more vulnerable to infection. As the contagion rate increases (γ), the set of vulnerable agents also increase.

We can also find the marginal probability of infection for each individual agents. Light yellow corresponds to low probabilty of infection whereas dark blue corresponds to high probability of infection. As the contagion rate increases (γ), the probability of infection increases for all the agents. We showed in analysis that when the contagion rate is low, agents with more neighbors have larger marginal probability of infection. When the contagion rate is high, agents belonging to densely connected communities have larger marginal probability of infection.



J. Zhang and J.M.F. Moura, “Cascading Edge Failures: A Dynamic Network Process,” submitted.


J. Zhang and J.M.F. Moura, “Contact process with exogenous infection and the scaled SIS process,” submitted.


J. Zhang and J.M.F. Moura, “Roles of subgraphs in network epidemics under the scaled SIS process," Journal of Complex Networks, 2015.

Abstract-In the previous work, we developed the scaled SIS process, which models the dynamics of SIS epidemics over networks. We derived for the scaled SIS process a closed-form expression for the time-asymptotic probability distribution of the configurations of all the agents in the network, which explicitly exhibits the underlying network topology through its adjacency matrix. This is accomplished for networks that are of finite-size and of arbitrary topology. This paper determines which network configuration is the most probable. We prove that, for a range of epidemic parameters, this combinatorial problem leads to a submodular optimization problem, which is exactly solvable in polynomial time. We relate the most-probable configuration to the network structure, in particular, to the existence of high-density subgraphs. Depending on the model parameters, subset of agents may be more likely to be infected than others; these more vulnerable agents form subgraphs that are denser than the overall network. We illustrate our results with a 193 node social network and the 4941 node Western US power grid under different model parameters.


J. Zhang and J.M.F. Moura, “Diffusion in social networks as SIS epidemics: beyond full mixing and complete graphs,” IEEE Journal of Selected Topics Signal Processing on Social Networks, 2014.

Abstract-Peer influence and interactions between agents in a population give rise to complex, nonlinear behaviors. This paper adopts the SIS (susceptible-infected-susceptible) framework from epidemiology to analytically study how network topology affects the diffusion of ideas/opinions/beliefs/innovations in social networks. We introduce the scaled SIS process, which models peer influence as neighbor-to-neighbor infections. We model the scaled SIS process as a continuous-time Markov process and derive for this process its closed form equilibrium distribution. The adjacency matrix that describes the underlying social network is explicitly reflected in this distribution. The paper shows that interesting population asymptotic behaviors occur for scenarios where the individual tendencies of each agent oppose peer influences. Specifically, we determine how the most probable configuration of agent states (i.e., the population configuration with maximum equilibrium distribution) depends on both model parameters and network topology. We show that, for certain regions of the parameter space, this and related issues reduce to standard graph questions like the maximum independent set problem.



J. Zhang and J.M.F. Moura, “Finding Unique Dense Communities,” in Proc. of the 41st IEEE International Conferences on Acoustics, Speech, and Signal Processing (ICASSP), Shanghai, China, March 2016, to be published.

To be published.

J. Mohammadi, J. Zhang, S. Kar, G. Hug, J.M.F. Moura, “Multilevel distributed approach for DC optimal power flow,” in Proc. of the 3rd IEEE Global Conference on Signal and Information Processing (GlobalSIP), Orlando, FL, Dec. 2015.

Abstract-The interest in distributed control methods for power systems is motivated by the need for scalable solutions to handle the coordination of an increasing number of distributed resources. This paper presents a fully distributed multilevel method to solve the DC Optimal Power Flow problem (DC- OPF). Our proposed approach constitutes a distributed iterative mechanism to solve the first order optimality conditions of the DC-OPF problem using the fact that optimality conditions involve local variable couplings. The proposed distributed structure requires each bus to update a few local variables and exchange information with neighboring buses. Our multilevel distributed approach distributes the computation at several levels, i.e., nodes, subareas and areas. It allows for synchronous information exchanges, i.e., after each iteration, at the nodal level and asyn- chronous communication, i.e., after multiple iterations, between subareas and areas. To define meaningful subareas, we are using a graph theoretic partitioning method derived from an epidemics model. We compare the performance of the proposed partitioning method over a random partitioning method using the IEEE 118- bus system.


J. Zhang and J.M.F. Moura, “Dynamic bond percolation in networks,” in Proc. of the 2nd IEEE Global Conference on Signal and Information Processing (GlobalSIP), Atlanta, GA, Dec. 2014.

Abstract-Bond percolation is a network process that traditionally addresses the question when there is a path between two sites or two clusters in a network. This has been used to study the circulation of goods or flows in networked structures as well as network resilience. This paper proposes and analyzes a dynamic bond percolation model where bonds (i.e., edges) open (i.e., form) or close (i.e., terminate) according to random micro interactions. We model the edge dynamics through topology independent and topology-dependent processes—spontaneous formation or termination of edges and the formation of edge {a, b} between two sites a and b that depends on the current number of existing edges at a and b. We show that the resulting network process is Markov and reversible, and we determine analytically its equilibrium distribution, avoiding having to solve a eigenvalue- eigenvector problem that quickly becomes intractable for even moderate sized networks.

J. Zhang and J.M.F. Moura, “Subgraph density and epidemics over networks,” in Proc. of the 39th IEEE International Conferences on Acoustics, Speech, and Signal Processing (ICASSP), Florence, Italy, May 2014.

Abstract-We model a SIS (susceptible-infected-susceptible) epidemics over a static, finite-sized network as a continuous-time Markov process using the scaled SIS epidemics model. In our previous work, we derived the closed form description of the equilibrium distribution that explicitly accounts for the network topology and showed that the most probable equilibrium state demonstrates threshold behavior. In this paper, we will show how subgraph structures in the network topology impact the most probable state of the long run behavior of a SIS epidemics (i.e., stochastic diffusion process) over any static, finite-sized, network..


J. Zhang and J.M.F. Moura, “Threshold behavior of epidemics in regular networks,” in Proc. of the 38th IEEE International Conferences on Acoustics, Speech, and Signal Processing (ICASSP), Vancouver, Canada, May 2013, pp. 5411-5414.

Abstract-Current research is interested in identifying how topology impacts epidemics in networks. In this paper, we model SIS (susceptible- infected-susceptible) epidemics as a continuous-time Markov process and for which we can obtain a closed form description of the equilibrium distribution. Such distribution describes the long-run behavior of the epidemics. The adjacency matrix of the network topology is reflected explicitly in the formulation of the equilibrium distribution. Secondly, we are interested in analyzing the model in the regime where the topology dependent infection process opposes the topology independent healing process. Specifically, how will network topology affect the most probable long-run network state? We show that for k-regular graph topologies, the most probable net- work state transitions from the state where everyone is healthy to one where everyone is infected at a threshold that depends on k but not on the size of the graph.

J. Zhang and J.M.F. Moura, “Epidemic process on fixed networks,” in 1st IEEE/ACM Workshop on Signal Processing Advancement in Sensor Networks, Philadelphia, USA, 2013.


J. Zhang and J.M.F. Moura, “Accounting for topology in spreading contagion in non-complete networks,” in Proc. of the 37th IEEE International Conferences on Acoustics, Speech, and Signal Processing (ICASSP), Kyoto, Japan, March 2012, pp. 2681-2684.

Abstract-e are interested in investigating the spread of contagion in a network, G, which describes the interactions between the agents in the system. The topology of this network is often neglected due to the assumption that each agent is connected with every other agents; this means that the network topology is a complete graph. While this allows for certain simplifications in the analysis, we fail to gain insight on the diffusion process for non-complete network topology. In this paper, we offer a continuous-time Markov chain infection model that explicitly accounts for the network topology, be it complete or non-complete. Although we characterize our process using parameters from epidemiology, our approach can be applied to many application domains. We will show how to generate the infinitesimal matrix that describes the evolution of this process for any topology. We also develop a general methodology to solve for the equilibrium distribution by considering symmetries in G. Our results show that network topologies have dramatic effect on the spread of infections.


J. Zhang, “LightCast: a tangible user interface creativity support tool for visual design,” in Proc. of 2006 ACM International Joint Conference on Pervasive and Ubiquitous Computing (UbiComp), Orange County, CA, 2006.


K U-Yen, M. Ahn, Z-J. Zhang, J.S. Kenney, “Effects of microwave switch isolation on a butler matrix beamforming network in smart antenna systems,” in Proc. of Radio and Wireless (RAW) Conference, Atlanta, GA, 2004.