Research
Grants and Awards
PI, "CRII: III: Federated Learning with Enhanced Efficiency and Privacy Preservation over Heterogeneous Devices", NSF, 7/2024 - 6/2026.
Co-PI, "Empowering Dementia Informal Caregivers in Managing behavioral symptoms with Generative AI Voice Assistant Education and Guidance App: A Pilot Feasibility Study", Kennesaw State University, Modupe Adewuyi (PI), 5/2024 - 4/2025.
Summer Research Fellowship, Kennesaw State University, 05/2024 - 06/2024.
PI, "First-Year Scholars Program: Leveraging Generative Models to Protect Machine Learning Frameworks", Kennesaw State University, 10/2023 - 5/2024.
Summer Research Fellowship, Kennesaw State University, 05/2023 - 06/2023.
PI, "First-Year Scholars Program: Two-Sided eCommerce Markets Fairness via Decentralized Matching Algorithm", Kennesaw State University, 10/2022 - 5/2023.
PI, ICWD SunTrust Fellow, "Secure and Private Deep Learning Models against Inference Attacks on the Internet of Things", Kennesaw State University, 8/2022-5/2023.
PI, "First-Year Scholars Program: Keep Your Data and Take Surveys: Secure and Private Data Aggregation", Kennesaw State University, 10/2021 - 5/2022.
Student Grant Award, IEEE International Conference on Communication (ICC), 2020
Student Travel Grant Award, IEEE International Conference on Distributed Computing Systems (ICDCS), 2019
Research Projects
Client Selection in Federated Learning: A Dynamic Matching-Based Incentive Mechanism
Federated learning (FL) has rapidly evolved as a distributed learning paradigm, enabling clients to collaboratively train models while retaining data privacy on their devices, which can guarantee the privacy of the training data. However, it faces distinct challenges on both server and client fronts. On the server side, there is a lack of efficient strategies for selecting high-performing clients, leading to potential degradation in training accuracy due to subpar model updates. On the client's side, they are often deterred from participation due to significant energy consumption during both computation and data transmission processes. Existing incentive mechanisms in FL seldom consider both the energy consumption of the clients and the learning quality of the server. To bridge this gap, this project introduces an adaptive incentive mechanism, which considers both the anticipated learning quality of clients and the associated energy costs during training. We propose a novel distributed Matching-based Incentive Mechanism (MAAIM) for client selection in FL. Leveraging a deferred acceptance algorithm, MAAIM facilitates stable client-server pairings, ensuring that both parties' primary concerns are addressed. Experimental results demonstrate the effectiveness of the proposed MAAIM.
Harnessing Inherent Noises for Privacy Preservation in Quantum Machine Learning
Quantum computing revolutionizes the way of solving complex problems and handling vast datasets, which shows great potential to accelerate the machine learning process. However, data leakage in quantum machine learning (QML) may present privacy risks. Although differential privacy (DP), which protects privacy through the injection of artificial noise, is a well-established approach, its application in the QML domain remains under-explored. In this project, we propose to harness inherent quantum noises to protect data privacy in QML. Especially, considering the Noisy Intermediate-Scale Quantum (NISQ) devices, we leverage the unavoidable shot noise and incoherent noise in quantum computing to preserve the privacy of QML models for binary classification. We mathematically analyze that the gradient of quantum circuit parameters in QML satisfies a Gaussian distribution, and derive the upper and lower bounds on its variance, which can potentially provide the DP guarantee. Through simulations, we show that a target privacy protection level can be achieved by running the quantum circuit a different number of times.
Energy-Efficient Federated Learning over Cooperative Relay-Assisted Wireless Networks
Federated learning (FL) is a promising distributed learning paradigm, which can effectively avoid the privacy leakage and communication issues compared with the centralized learning. Specifically, in each training iteration, FL nodes only upload the local training results to the centralized server without disclosure of their raw training dataset and the centralized server will aggregate the local results of all FL nodes and update the global model. To this end, the performance of global model is highly dependent on the nodes' cooperation. However, it is challenging to motivate mobile edge devices to involve in the FL process without a desired incentive. Another significant concern of mobile edge devices is the communication and computation energy cost of the participation. Therefore, considering the high cost and weak communication channel with the centralized server specially for the distant nodes, in this paper, we propose a relay-assisted energy efficient scheme for federated learning, where each FL computational node is not only motivated by monetary awards based on their local dataset, but also further motivated to function as a relay node to assist distant nodes on local results uploading due to its locality advantage. We propose two methods to solve the relay selection problem: (1) matching algorithm; (2) reinforcement learning (RL). To achieve a stable pairing solution between FL computational nodes and assisted relays in a distributive fashion, a many-to-one matching algorithm is applied, where each the computational node and relay is unable to deviate with current pairing unilaterally for higher revenue. We also propose an RL model to tackle the node-relay selection problem. The model we have developed uses a deep Q-network (DQN) algorithm, a deep learning (DL) variation on the classic Q-Learning, to train our model’s agent. Our model, which we call the DQN-Relay model, consists of two DQNs, a main network and a target network, to balance training. Extensive simulations are conducted to illustrate the correctness and effectiveness of our proposed scheme.
Differential Privacy in Generative Adversarial Networks
In recent years, generative adversarial network (GAN) has attracted great attention due to its impressive performance and potential numerous applications, such as data augmentation, real-like image synthesis, image compression improvement, etc. The generator in GAN learns the density of the distribution from real data in order to generate high fidelity fake samples from latent space and deceive the discriminator. Despite its advantages, GAN can easily memorize training samples because of the high model complexity of deep neural networks. Thus, training a GAN with sensitive or private data samples may compromise the privacy of training data. To address this privacy issue, we propose a novel Privacy Preserving Generative Adversarial Network (PPGAN) that perturbs the objective function of discriminator by injecting Laplace noises based on functional mechanism to guarantee the differential privacy of training data. Since generator training is considered as a post-processing step while guaranteeing differential privacy of discriminator, the trained generator should be differentially private to effectively protect data samples. Through detailed privacy analysis, we theoretically prove that PPGAN can provide such strict differential privacy guarantee.
Data Privacy Preserving for Cyber-Physical Systems
Cyber-physical systems (CPS) often referred as “next generation of engineered systems” are sensing and communication systems that offer tight integration of computation and networking capabilities to monitor and control entities in the physical world. The advent of cloud computing technologies, artificial intelligence, and machine learning models has extensively contributed to these multidimensional and complex systems by facilitating a systematic transformation of massive data into information. Though CPS have infiltrated into many areas due to their advantages, big data analytics and privacy are major considerations for building efficient and high-confidence CPS. Many domains of CPS, such as smart metering, intelligent transportation, health care, sensor/data aggregation, crowdsensing, etc., typically collect huge amounts of data for decision-making, where the data may include individual or sensitive information. Since vast amount of information is analyzed, released, and calculated by the system to make smart decisions, big data plays a key role as an advanced analysis technique providing more efficient and complete solutions for CPS. However, data privacy breaches during any stage of these largescale systems, either during collection or big data analysis, can be an undesirable loss of privacy for the participants and for the entire system.
Location Privacy in Transportation Network Company Vehicle Scheduling
With the popularity of mobile devices with global positioning system (GPS), transportation network company (TNC) service has become an indispensable option of people's daily commute. However, it also provides opportunities for malicious parties to compromise TNC users' location privacy. There are great challenges to preserve TNC users' location privacy while improving the revenue of TNC and its quality of service (QoS). To address this issue, we propose a novel scheme to schedule the TNC vehicles while preserving the TNC users' location differential privacy. Briefly, we add high dimensional Laplace noises to guarantee the TNC users' geo-indistinguishability. Due to the differential private obfuscation, the demand for TNC vehicles in an area becomes uncertain. Thus, we employ the data-driven approach to characterize users' demand uncertainty, formulate the TNC's revenue maximization problem into risk-averse stochastic programming, and provide corresponding feasible solutions. Using the released public data of Didi Chuxing, we evaluate the performance of the proposed scheduling scheme and compare the results under different zeta-structure metrics. The results show that the proposed scheme can efficiently schedule the TNC vehicles, maximize the TNC's revenue and provide a better service for TNC users while protecting the TNC users' location privacy.
Differential Privacy in Mobile Crowdsourcing
Mobile crowdsensing (MCS) has become a new sensing and computing paradigm due to the proliferation of global positioning system (GPS) enabled mobile devices. There are three parties in the MCS, the MCS server, task requesters and workers. The MCS server needs to collect workers' location information to optimize the task allocation problem. However, during the location data collection process, workers' location privacy might be disclosed without their knowledge. It is challenging to preserve workers' location privacy while effectively and efficiently selecting proper workers to fulfill an MCS task. We propose a novel differentially private geocoding (DPG) mechanism to preserve workers' location privacy. Specifically, instead of reporting the exact latitude and longitude to the server, workers can use obfuscated geocode to describe their locations, since geocodes can provide an intuitive visualization of workers' spatial information to the MCS server. Based on the workers' obfuscated geocodes, we also formulate a travel distance minimization problem in MCS into an integer linear programming problem. We leverage conditional value at risk (CVaR) to characterize the uncertainty brought by the obfuscated geocodes, and develop feasible solutions to the formulated optimization problem.
Google Scholar: Link