I am an Insight data-science fellow in Boston, MA, where I have developed a natural language processing (NLP) pipeline to predict public-favorability of self-driving cars based on Twitter feed. Prior to that, I did my PhD and postdoctoral fellowship in the department of Electrical and Computer Engineering at the University of Waterloo, Canada, with a focus on modeling vehicle mobility and its effect on vehicular communication networks. During my PhD, I built a stochastic model that describes driver behavior and its affect on the collective movement of vehicles on highways, enabling scalable probabilistic characterization of temporal variabilities (vehicle clusters, network topology, and protocol performance) in a network of wirelessly connected vehicles. During my postdoctoral fellowship, I was engaged in projects focused on exploiting WiFi (DSRC IEEE 802.11 p) and cellular (LTE) technologies for both human-driven and autonomous vehicles to enhance on-road safety. For more about my research please check my research portfolio below.
Throughout my graduate studies and postdoctoral fellowships, I have developed expertise in stochastic graphical modeling, survival analysis, and machine learning. I am interested in understanding and modeling human behavior and leveraging it in designing solutions that optimize human experience.
Besides research, I enjoy volunteering and organizing social and networking activities, for over a year I organized the graduate research seminars at the ECE department, University of Waterloo, and the vehicular communications group in our Lab, I am currently serving as a fellowship co-chair on the board of ACM SIGMOBILE N2Women community, where I help organize networking sessions in international conferences. In my free-time, I enjoy running, power yoga, and exploying cities on foot (and I mean walking +8hrs/day). I appreciate mimimalism and idea-inducing art and conversations. At home, I very much enjoy roasting coffee beans and pouring latte-art. If you are interested in learning about how to roast coffee beans and make a delicious cup of coffee at home with little equipment, check out my personal section.
To accurately assess temporal variabilities in vehicular communication networks, a microscopic vehicle mobility model is required. Despite the rich literature in vehicle traffic flow modeling, off-the-shelf microscopic mobility models are designed with simulator-level sophistication that hinders network analysis. In my PhD, I have developed a novel stochastic generative model that describes temporal variations in vehicle movements enabling scalable and traceable mathematical characterization of temporal variations in vehicular networks.
I used a discrete-time finite-state Markov chain with state-dependent transition probabilities to model the distance headway between two consecutive vehicles. In my model, a driver behavior is abstracted in the temporal changes in distance headway, which are governed by the transition probabilities. I modeled the transition probabilities from each state as functions of the distance and the road traffic density, thus, implicitly accounting for the correlations existing in vehicle movement with nearby vehicles' behaviors. When the vehicle density is high, the state-dependency parameter is set to a large value, implying more interactions between vehicles since drivers become more cautious to their current distance.
I tuned the markov chain model using i) empirical vehicle trajectory data collected from two highways (I-80 and US-101 ) in the state of California and provided by the next generation simulation (NGSIM), a project by the United States department of transportation; and ii) simulated vehicle trajectory data that I generated by VISSIM microscopic vehicle traffic simulator, which is based on a sophisticated multi-modal psycho-physical mobility model that mimics realistic vehicular traffic. The Markov chain parameters, i.e., transition probabilities, are calculated using maximum likelihood estimation (MLE).
Furthermore, I extended my mobility model to a group mobility model that describes the time variations of a system of distance headways between two non-consecutive vehicles. Using lumpability theory, I derived a Markov chain with reduced state-space to represent the mobility of a group of vehicles. I mathematically proved the lumpability of the multi-dimensional Markov chain into my proposed state-space partition. The steps of modeling individual and group vehicle mobility is illustrated the figure below.
The lumped-group mobility model enables the exact stationary and survival (transient) analyses for a system of distance headways. Using the lumped Markov chain, the scalability of analyzing the movement of a group of vehicles with respect to the size of the group (i.e., the number of vehicles) is improved for both the steady-state and survival(transient) analysis. I proved this via computational complexity analysis (big O notation derivation) on the lumped Markov chain. The computational complexity of survival(transient) analysis are functions of the state-space of the considered Markov model. The state space reduction using my proposed lumped Markov chain, shown in the plots below, is able to tackle the main problem in probabilistic graphical models which is state-space expolsion, and the resulting lumped Markov model enables modeling and analyzing the collective sequential behavior of a system composed of independently behaving components.
Due to high and variable relative speeds among vehicles, a vehicular network is subject to frequent temporal variations in its topology. Moreover, vehicular networks are susceptible to vehicle density variations from time to time throughout the day. Such temporal topological variations can cause network fragmentation and interrupt on-going packet transmissions between two connected vehicles, potentially disrupting network applications and services for the driver or the passengers.
In this work, I used my stochastic microscopic vehicle mobility model (Section 1) to analyze and characterize the temporal variations in network topology. I proposed two characteristics to describe the impact of vehicle mobility on the vehicular network topology: 1) the time period between successive changes in communication link state (connection and disconnection), and 2) the time period between successive changes in vehicle's one-hop neighborhood. By applying survival(transient) analysis, both first-passage-time and absorption-time analyses, on my proposed lumped group mobility model, the two network topology characteristics were probabilistically described for different vehicular traffic flow conditions.
Additionally, using Queuing and Renewal theories, I modeled the system of two-hop vehicles and the common vehicle neighbors between them as a storage buffer with two overlap-state random environment. I used diffusion approximation to describe the limiting behavior of this system and approximate it by a simple quantitative measure of the steady-state number of common vehicle neighbors between two-hop vehicles. The figure below illustrates the pipeline I developed for mathematically characterizing temporal network topology using my proposed vehicle mobility model. The probabilistic temporal topology characterization provides a useful tool for the development of mobility-aware vehicular network protocols.
Due to the high vehicle speeds and the frequent topology changes in vehicular communication networks, finding and maintaining routes are challenging tasks. Wireless links switch between connection and disconnections because of the relative speed between the nodes, thus increasing the routing overhead associated with topology updates and route discovery processes.
Node clustering is a network management approach that has been used to improve the scalability of routing protocols. Cluster-based routing protocols proposed in the literature aim to minimize the routing overhead and scale to an increased node density. In this work I have probabilistically analyzed the impact of steady-state cluster characteristics in terms of the cluster size and the cluster-overlap on reducing generic (reactive/proactive) routing overhead. In the second part of this work, I proposed a probabilistic characterization of the impact of cluster instability, due to vehicle mobility, on generic routing overhead. First, the temporal variations in network topology (discussed in Subsection2.1) were mapped to cluster-instability measures as illustrated in Figure 7. The cluster stability model characterizes the impact of microscopic vehicle mobility on the internal and external cluster stability. The probability distribution of the time between successive changes in cluster membership is used as a measure of internal cluster stability. External cluster stability is measured by the probability distributions of the number of common/unclustered nodes between overlapping/disjoint neighboring clusters. I derived the probability distribution of the intra- and inter-cluster routing overhead imposed by vehicle mobility taking into account the external cluster stability impact on route availability and cluster connectivity. Numerical results show that cluster instability increases intercluster routing overhead such that it becomes probabilistically more than that of a controlled flooding with a relay percentage 25%. The probability distributions derived in this work provide indicators for the impact of cluster instability on the routing overhead, which can be used to inform the design of mobility-aware cluster-based routing protocols for vehicular communication networks.
There are two existing and competing technologies that can potentially enable vehicle-to-everything (V2X) communications: 1) the dedicated short range communications (DSRC) which is based on the IEEE 802.11 p amendment of the legacy WiFi standard, and 2) cellular network technologies. However, in the near future, it is not expected that a single technology can support the variety of expected V2X applications for a large number of vehicles. High vehicle density increases packet collisions in contention-based medium access in DSRC (governed by the fundamental CSMA/CA technique) and creates a bottleneck in pure-cellular based network. Hence, interworking between DSRC and cellular network technologies for efficient V2X communications is proposed. In this work, I surveyed potential DSRC and cellular interworking solutions to support V2X communications. First, I highlight the limitations of each technology in supporting V2X applications. Then, I reviewed potential DSRC-cellular hybrid architectures, together with the main interworking challenges resulting from vehicle mobility, such as vertical handover and network selection issues. In addition, my survey provides an overview of the global DSRC standards, the existing V2X R&D platforms, and the V2X products already adopted and deployed in vehicles by car manufactures, as an attempt to align academic research with automotive industrial activities. Finally, I recommend some open research issues for future V2X communications based on interworking of DSRC and cellular network technologies.
Furthermore, I proposed a hybrid architecture leveraging the interworking of cellular and DSRC technologies to support reliable V2X communications for event-driven safety applications as illustrated in the figure below. My proposed solution avoids limitations of each of the technologies and utilizes their strengths. Particularly, I leverage evolved Multimedia Broadcast Multicast Services (eMBMS), which are available in LTE to support reliable and low latency event-driven safety message broadcast in the emergency zone, that is generally larger than the DSRC communication range. I identify the methods of calculating the event-driven safety application QoS requirements for each of the two networks.
Driver error is the largest contributor to motor vehicle crashes in the U.S with an estimated 94% (±2.2%) of the crashes attributed to driver error according to the latest NHTSA crash assessment report (NHTSA Rep. DOT HS812013, 2015). The wide adoption of self-driving cars could dramatically reduce the number of traffic accidents. However, many hurdles still face the full deployment of self-driving cars, from regulatory to technical issues. One of the major challenges, is public opinion. A recent survey has shown that 75% of people are very/moderately/slightly concerned about self-driving cars, which re ects the level of skepticism towards the technology (Kyriakidis, M. et al. 2015).
As an Insight data-science fellow, I tried to answer the following question: Can the automotive industry engage in a different conversation with the public such that it sways its perception of self-driving cars to a more favorable one? To this end, I have developed an NLP pipeline that predicts the public-favorability of self-driving cars using Twitter feed. I collected twitter feed related to self-driving cars using the Twitter REST API, and filtered the tweets to those published by verified accounts, which have the most influence on public opinion. I pre-processed the tweets through text normalization (lemmatization and cleaning from URL, mentions, and stop words). Using a bag-of-words model and term frequencyinverse document frequency (TF-IDF) weighting, the pre-processed text data is transformed to a vectorized feature matrix. I have also included the compound sentiment of the tweet as a feature using valence aware dictionary for sentiment reasoning (VADER), which is a rule-based sentiment analysis tool pre-trained on twitter data (Hutto, Clayton J. et al. 2014). The pipeline for feature extraction is illustrated the figure below.
Using the tweets' attributes (the number of favorites (i.e., likes) and the number of followers of the publisher of the tweet), I have labeled the tweets into favorable/non-favorable. The favorability of a tweet is calculated by a function of the number of favorites normalized to the number of followers. Then, I have trained and cross-validated a regularized logistic regression with L1- penalty to predict whether twitter users would favor a tweet related to self-driving cars. I have also identified the features (words and tweet sentiment) with the highest predictive power.
The trained model has 67.5% accuracy evaluated on hold-out test data. The pipeline for predicting the favorability of self-driving tweets is illustrated in the figure below. Despite using the tweet's sentiment as a feature in my model, it had small predictive power as compared to the tweet's text. This is because the tweet's sentiment reflects the publisher's opinion rather than the public's perception of the tweet, which shows that using an off-the-shelf sentiment analysis tool is not effective in predicting the public's opinion about self-driving cars. I have deployed my trained model in a web application (AUTOMOTivate), where an automotive company can check their tweet before publishing it and iterate through it according to the recommended words. My application aims to help automotive industry advertise their self-driving car technology through social media and increase their favorability in public opinion.
K. Abboud and W. Zhuang, Mobility Modeling for Vehicular Communication Networks. Springer Briefs in Electrical and Computer Engineering, 2015, ISBN 978-3-319-25507-1 DOI: 10.1007/978-3-319-25507-1
H. Peng, D. Li, Q. Ye, K. Abboud, H. Zhao, W. Zhuang, and X. Shen. "Resource allocation for cellular-based inter-vehicle communications in autonomous multiplatoons," IEEE Transaction on Vehicular Technology, accepted on June 30th, 2017.
K. Abboud, H. Omar, and W. Zhuang, "Interworking of DSRC and cellular network technologies for V2X communications: A survey," IEEE Transactions. Vehicular Technology vol. 65, no.12, pp. 9457 - 9470, 2016.
K. Abboud and W. Zhuang, "Stochastic modeling of single-hop cluster stability in vehicular ad hoc networks," IEEE Transactions on Vehicular Technology, vol. 65, no. 1, pp. 226 - 240, 2016.
H. Omar, K. Abboud, N. Cheng, K. Rahimi Malekshan, A. T. Gamage, and W. Zhuang "A survey on high efficiency wireless local area networks: Next generation WiFi," IEEE Communications Surveys and Tutorials, vol.18 no.4, pp. 2315 - 2344, 2016
H. Peng, D. Liz, K. Abboud, H. Zhou, W. Zhuang, X. Shen, and H. Zhao, "Performance analysis of IEEE 802.11p DCF for multiplatooning communications with autonomous vehicles,"IEEE Transactions. Vehicular Technology, vol. 66, no. 3, pp. 2485-2498. 2017.
K. Abboud and W. Zhuang, "Impact of microscopic vehicle mobility on cluster-based routing overhead in VANETs," IEEE Transactions. Vehicular Technology, Special Series on Connected Vehicles, vol. 64, no. 12, pp. 5493 - 5502, 2015.
K. Abboud and W. Zhuang, "Stochastic Analysis of Single-Hop Communication Link in Vehicular Ad Hoc Networks," IEEE Transactions on Intelligent transportation systems, vol. 15, no. 5, pp 2297 - 2307, May 2014.
H. Peng, D. Liz, Q. Ye, K. Abboud, H. Zhou, W. Zhuang, X. Shen, and H. Zhao, "Resource allocation for D2D-enabled inter-vehicle communications in multiplatoons". Proc. IEEE ICC, 1-6, 2017.
H. Peng, D. Liz, K. Abboud, H. Zhou, W. Zhuang, X. Shen, and H. Zhao, "Performance Analysis of IEEE 802.11p DCF for Inter-platoon Communications with Autonomous Vehicles," Proc. IEEE Globecom, San Diego, CA, USA, Dec. 2015.
K. Abboud and W. Zhuang, "Impact of Node Mobility on Single-Hop Cluster Overlap in Vehicular Ad Hoc Networks," Proc. ACM MSWiM, Montreal, QC, Canada, Sept. 2014. (reveived the Best Paper Award).
K. Abboud and W. Zhuang, "Analysis of Communication Link Lifetime Using Stochastic Microscopic Vehicular Mobility Model," Proc. IEEE Globecom, Atlanta, GA, USA, Dec. 2013.
K. Abboud and W. Zhuang, "Impact of Node Clustering on Routing Overhead in Wireless Networks," Proc. IEEE Globecom, Houston, TX, USA, Dec. 2011.
K. Abboud and W. Zhuang, "Modeling and Analysis for Emergency Messaging Delay in Vehicular Ad Hoc Networks," Proc. IEEE Globecom, Honolulu, HI, USA, Dec. 2009.
W. Zhuang and K. Abboud, "Modeling and analysis of vehicle mobility for clustered vehicular ad hoc networks," Ontario and Canada Research Chairs (CRC) Symposium: Research Matters, Toronto, ON, Canada, April 2015.
K. Abboud and W. Zhuang. "Cellular-DSRC hybrid architecture for V2X communications" In Fifth Networking Networking Women Workshop co-located with International ACM MobiCom Conference, Manhattan, NY, USA, 2016
Fellowship co-chair for Networking Networking Women (N2Women) (ACM SIGMOBILE), 2016-2018
Grace Hopper Celebration (GHC) of Women in Computing, 2017
IEEE Vehicular Technology Conference (VTC-Fall), 2016 - 2017
IEEE Global Communications Conference (Globecom), 2015 - 2016
IEEE International Conference on Scalable Computing and Communications (ScalCom), 2014
International Conference on Communications Engineering (ICComE), 2010
"Wireless Networks (II)" Symposium (D3-S1) of the 17th ACM International Conference on Modeling, Analysis and Simulation of Wireless and Mobile Systems (MSWiM), Montreal, Canada, 2014
"Query Patterns and Sensor Networks" 4th ACM International Symposium on Design and Analysis of Intelligent Vehicular Networks and Applications (DIVANet), Montreal, Canada, 2014
"Smart Cars and Social Network" 4th ACM International Symposium on Design and Analysis of Intelligent Vehicular Networks and Applications (DIVANet), Montreal, Canada, 2014
"Topics in Ad Hoc and Sensor Networks in the Ad Hoc and Sensor Networking" Symposium (AHSN-19) of the IEEE Global Communications Conference (Globecom) Atlanta, USA, 2013
IEEE Transactions on Vehicular Technology (TVT), 2014 -2016
IEEE Vehicular Technology Conference (VTC), 2013, 2016
IEEE Transactions on Intelligent Transportation Systems (ITS), 2015
IEEE Vehicular Technology Magazine, 2015
IEEE international conference on communications (ICC), 2012, 2013, 2015
Electronics and Telecommunications Research institute Journal (ETRI), 2014, 2016
IEEE Global Communications Conference (Globecom), 2011, 2014
Journal of Circuits, Systems, and Computers (JCSC), World Scientific, 2013
International Symposium on Wireless Vehicular Communications (WIVEC), 2013
The International Journal for the Computer Communications, Elsevier, 2012
IEEE Vehicular Networking Conference (VNC), 2010
Organizer for vehicular ad hoc networks (VANETs) BBCR subgroup, 2015 - 2016 (University of Waterloo)
Volunteer in the IEEE 33rd International Conference on Computer Communications (INFOCOM), 2014 (Toronto, ON, Canada)
Organizer in TEDxWaterloo production team, 2012 (Kitchener, ON, Canada)
Participant in graduate course critique screening for Faculty of Engineering, 2013 (University of Waterloo)
Arabic language tutor or UWAT program, 2012 (University of Waterloo)
Host for the United Nation Arabic Language day, 2012 (University of Waterloo)
Volunteer in TEDxWaterloo - day of volunteer (DOV), 2011 (Kitchener, ON, Canada)
Organizer for Graduate Research Seminars in Electrical and Computer Engineering Department, 2008 - 2009 (University of Waterloo)
Vice chairman of IEEE student chapter, Kuwait University Branch, 2005-2007 (Kuwait University)