Sample Research Projects

and the Department of Defense (DoD)

The project REU participants are working to catalyze a core multidisciplinary research team consisting of senior and early career researchers, and explore theoretical, design and software issues for next generation technologies. The projects augument core research themes missing from current solutions' optimization methods and software for robust, secure, context aware, reconfigurable and adaptable computing systems for emerging multidisciplinary applications. This integrated need is expected to help REU students understand the context of interdisciplinary research and interact with other students in the group. Since computer science, communications and computing fields evolve rapidly, specific projects offer during the course of the program. Some of the projects currently underway (reflecting the types of the projects that are available to REU participants) are listed below along with projects that past participants have worked on.

If you are interested in working on a project, please contact the faculty listed as mentors. You may also review projects conducted by the KFSCIS Faculty and ECE Faculty and contact Dr. Pissinou if you wish to be matched with one of these professors.

Improve Robustness of Sensor Network Machine Learning Models Against Adversarial Attacks

In the field of adversarial machine learning (ML), an adversary seeks to exploit ML models through malicious input either to cause privacy violations or misprediction through adversarial examples. Various works have focused on securing classification problems, such as with image misclassification or spam email filtering. Minimal research has focused on adversarial attacks against multivariate time series models, as is seen in sensor networks, due to the added complexity introduced by the temporal relationship and additional challenges added with evaluating the efficacy of regression attacks. This research will introduce a novel adversarial attack algorithm that will exploit the temporal patterns of data to maximize model output error while minimizing potential detectable adversarial examples. REU students may participate in: (1) identifying an effective loss function for spatiotemporal datasets to create the highest quality adversarial examples; (2) aid in improving the complexity of the algorithm such that it is more accessible to use in resource-constrained sensor network applications such as in mobile wireless sensor networks; and (3) simulate the adversarial attacks over a variety of machine learning models and datasets.

Machine Learning Models for Sensor Network Applications Using Quantum Principles

Deep learning has been shown to be incredibly vulnerable to hard-to-detect adversarial examples. However, it is increasingly being used to aid in the data cleaning and feature selection in applications with highly complex data, such as in sensor network applications. To explore the fundamental characteristics that contribute to the learning models' sensitivity to adversarial examples, this research will investigate how non-linearity in activation functions influences robustness against adversarial attacks for secure sensing applications. We plan to study this by introducing a new class of highly non-linear activation functions using quantum mechanics principles. REU students may participate in: (1) deriving the quantum mechanics principles into a set of functions that can be used as activation function for sensing applications; (2) simulating the adversarial attacks over a variety of machine learning models and datasets; (3) calibrating and verifying the impact on the robustness of the newly introduced parameters by quantum mechanics equations; and (4) evaluate the computational complexity of the new activation functions to ensure it is accessible to use in resource-constrained sensor network applications such as in mobile wireless sensor networks.

Blockchain-based Sensor Network Applications with Split-Merge Problems due to Node Mobility

Blockchain technologies are maturing at a rapid pace and being used in a wide range of real-world applications. Traditional blockchain systems are designed in a linear-based structure which can lead to poor scalability, throughput, and conformation. Although several works were proposed to address these performance bottlenecks using linear and graph-based blockchains, they still do not provide robust solutions to address some key challenges in mobile networks, such as mobility, group splitting and merging, and maintaining trustworthiness when groups of nodes split and merge. REU students would build upon a successful and well-received framework developed by previous REU students. This cohort of REU students would be working on: (1) developing a new graph-based blockchain structure (named Merkle DAG), instead of traditional linear-based structures, that can facilitate the split and merge of multiple blockchains, (2) allowing multiple networks or blockchains to work independently while maintaining trust in a trust-less environment, and (3) designing a blockchain system that can improve the overall performance over traditional blockchain systems in terms of confirmation time, throughput, scalability, and addressing the critical challenges associated with mobile IoT systems, as mentioned earlier.

Data Cleaning Li-Fi Wireless Sensor Networks

WSNs, the central nervous system of IoTs, are placing an unprecedented demand on the wireless spectrum, making it scarcer and less secure. Emerging Light-Fidelity (Li-Fi) technology promises to alleviate these concerns while increasing both speed and security. The faster transmission speeds increase the amount and types of data that need to be filtered, merged, compared, contrasted, interpolated, and extrapolated, thus emerging as a new challenge in data analytics. Over the past few years, there has been a surge of interest from REU students in working on data cleaning problems for WSNs by applying new abstractions and statistical techniques. Moving forward, REU students may explore qualitative data cleaning approaches for Li-Fi by (1) developing Li-Fi-based WSN testbeds to explore how data cleaning approaches work and can be improved; (2) identifying data quality problems for semi-structured and unstructured data; and (3) developing qualitative data cleaning approaches for distributed streams of data.

Trajectory Privacy Preservation in Internet of Things

Trajectory information indicates the movement patterns of sensors, and reveals the personal preferences and habits of users. Trajectory privacy invasion could lead to several types of risks for users, including personal safety, profiling, and interference attacks. This research aims to answer three interrelated questions: (1) what form of trajectory information could be invaded by attackers without accessing the database and how does it happen? (2) What methods can be used to protect trajectory privacy from both external and internal attacks undergoing data transmissions? and (3) What are the factors critical to building a trajectory privacy attack-tolerant localization system? The answers will result in a novel trajectory privacy-aware localization system, involving multiple, potentially selfish, parties. REU students may participate in (1) simulating a new privacy preserving routing protocol; (2) developing the software to implement the new routing protocols into memsic sensor boards; (3) conducting static wireless sensor network experiments for trajectory data logging; (4) modifying, calibrating, and verifying the sensor kits to properly operate with sensor mobility; and (5) collecting and analyzing full-scale real data in the experiment fields.

Security and Privacy of Flying Ad Hoc Networks

In an era where companies are planning to deliver customer orders using drones, it is important to ascertain the need for security and privacy of the drones. Although a network of drones can be modeled as a Flying Ad Hoc Network (FANET), existing ad hoc network protocols do not perform adequately in FANETs due to the high degree of node mobility. With limited airspace this can thus result in mid-air collisions as well as breaches in security and privacy. This project examines new security and privacy protocols for drones in FANETs and using location-based services. REU students may participate in (1) developing security protocols for high mobility degree nodes; (2) designing a fast and secure group key protocol, forming groups inside FANETs; (3) developing new techniques for using location-based services for air traffic; and (4) designing a privacy-preserving location assurance protocol for location-aware services in FANETs.

Secure and Trusted Crowdsourcing Schemes in Mobile Sensor Networks

Applying crowdsourcing principles to mobile sensor networks to perform tasks with human involvement, collaborative real-time collection and processing could ensure the efficiency and robustness of data processing, while limiting the required communication bandwidth. Mobile crowdsourcing networks (MCNs) face many potential attacks, such as reward forging, and collusion attacks. A key issue is how to prevent malicious colluding participants from generating false consensus regarding an event and prevent compromised participants from generating false observations. The goal of this project is to significantly increase the complexity and cost of colluding attacks that seek to influence consensus of observations or generate false or repeated data in mobile sensor networks. REU students will (1) develop 8 crowdsourcing schemes that detect and resist forging and collusion attacks; (2) develop privacy-preserving truth discovery methods that explore the intrinsic properties between data reports; and (3) develop trust modeling frameworks based on behavioral game theory.

ICN-Enabled Secure Edge Networking with Augmented Reality

This project aims to drive a new wireless edge network architecture design integrated with augmented reality (AR) mechanisms, through the exercise of two distinct application scenarios: AR-enriched campus daily life and disaster recovery. We propose a design that would enable intrinsic security, scalable content caching and discovery, in-network hardware acceleration, and unique time-saving features enabled by NDN’s direct use of application data names in the network layer. To successfully enable AR at the wireless edge, both communication and computation functions must be localized at the edge, thus eliminating long delays accessing cloud services, and improving users’ quality of experience. Undergraduates may participate in (1) designing a consistent naming system for content produced by all cyberspace entities (sensors, cameras, mobile devices, and users); (2) developing a prototype AR system that leverages NDNbased communication to retrieve data using any available channels, avoiding dependency on cloud infrastructure; and (3) performing simulation-based studies to measure communication quality.

Designing an Energy-Efficient Data Harvesting Framework using NDN

One of the key characteristics of wireless sensing networks (WSN) is limited power resources. Therefore, it is vital to conserve power as much as possible, while efficiently and timely retrieving the captured data. Named data networking (NDN) architecture provides unique opportunities for WSNs to save energy while providing high efficiency for data retrieval. In this project REU students will (1) extend the existing work in this area by designing an analytical model for joint optimization of data retrieval strategy, cooperative data storage, and sleep cycle synchronization, and (2) evaluate the developed model using a simulationbased study using an ndnSIM module for the NS-3 network simulator, gaining insights into the analytical model and validating its properties.

Camera-Based Two-Factor Authentication

Mobile and wearable devices are popular platforms for accessing sensitive online services such as ecommerce, e-mail, social networks, and banking. A secure and practical user authentication approach in such devices is challenging, as their small form complicates the input of commonly used text-based 6 passwords, even as the memorability of passwords already poses a significant burden for users trying to access a multitude of services. Furthermore, while biometric authentication provides sufficiently strong security, biometrics are hard to keep secret and pose lifelong security risks to users when stolen, as they cannot be reset and re-issued. More importantly, as surrendering biometrics may become mandatory, existing vulnerabilities, coupled with the compromise of large-scale biometrics databases, raise significant long-term security concerns. Alternatives such as a token-based authentication solution require expensive infrastructure. In this project, REU students will develop a camera-based remote authentication solution for mobile devices. They will participate in (1) designing the details of the solution; (2) implementing a prototype; and (3) conducting a pilot field study to evaluate the adoptability of the solution into the user’s daily life.

Security Parameters Bootstrapping in WSN Named Data Networking

One embedded feature of NDN is application-named data with built-in security, multicasting, and innetwork storage, which improves efficiency of WSNs. However, the basic communication primitive of NDN—fetching named and secured data packets—depends on proper security configuration of individual sensor nodes and the system as whole. Each sensor node needs to be provisioned with cryptographic keys and certificates, so the data the node generates are trusted by other nodes. In this project, the REU students will (1) develop method(s) to effectively and securely bootstrap and maintain security credentials of the deployed WSN without requiring physical access to the nodes; (2) develop trust and privacy mechanisms for NDN architectures that satisfy the complex interaction among IoT devices and mobile wireless sensor networks; and (3) analyze attack models and defense strategies for NDN-based WSNs.

Erasure Coding with High Parallelism for Distributed Mass Storage Systems

It is critical for distributed mass storage systems to maintain data reliability when some storage devices are compromised due to physical or software failures. To make the distributed mass storage system resilient to such failures, it must store redundant data generated from original data and maintain data reliability, even if data on some of its storage devices are compromised. Traditional coding techniques limit the performance of data analytic jobs. We propose to design innovative coding techniques that can offer high parallelism for parallel data frameworks and still maintain high resilience to failures. REU students may participate in (1) creating novel designs of erasure codes to enable high parallelism in parallel data frameworks; (2) designing and implementing a distributed mass storage system based on Apache Hadoop; and (3) building a new parallel data framework based on Apache Spark.

Building and Managing Software-Defined Infrastructures for WSNs

The recently available Software Defined Networking (SDN) paradigm dramatically simplifies network configuration and operation and enables enforcement of fine-grained security policies by implementing a centralized, programmable control plane. Recently researchers started applying the SDN architecture to wireless sensor networks, resulting in software-defined WSN (SD-WSN). The development of SD-WSN, however, must address two critically important challenges—control traffic overhead and information correlation. Building on previous REU studies, students will focus on the specific domain of SD-WSN, and will (1) develop and implement an efficient routing algorithm to minimize the overhead of control traffic, by applying the design of MMC in the SD-WSN context; (2) design and implement control modules that automate management tasks such as monitoring and diagnosis, by applying standard correlation techniques such as Shared-Risk link group; and (3) design and implement proper visualizations of the monitoring and diagnosis information, using semantics-based grouping schemes and multi-level, interactive interface.

Cryptographic Primitives for MWSNs

Cryptographic primitives are fundamental building blocks for constructing security protocols to achieve confidentiality, authentication, and integrity. Given that many cryptographic primitives are based on the intractability of number-theoretic problems such as factoring and discrete logarithms, and are constructed based on unproven complexity hardness assumptions, it is desirable to consider WSN protocols based on subset-sum and decoding random linear code problems. Faster space-efficient algorithms for Subset-Sum and related NP-hard problems could combine previously developed techniques or be applied to solve the learning parity with noise problem, tackle some NP-hard problems and achieve polynomial speed-ups. For example, in we introduced a novel idea based on error correcting code to obtain very simple and fast algorithms for the closest pair of vectors problem. We intend to explore whether the idea can be generalized to other search problems. Specific tasks for REU students will be to (1) understand the ideas and analysis of the state-of-the-art algorithms for some NP-hard problems; (2) implement one or more of these algorithms, comparing the performance of different algorithms on a small testing data set; and (3) brainstorm new ideas for the design and analysis of new, improved algorithms for these NP-hard problems.