Click on any of the track links below for more info on each talk. All listed tracks and keynotes take place Friday, October 25, 2024.
Keynote Talks
- James M. Taylor, Jr., Nebraska Defense Research Corporation, "Rise of the Machines? Opportunities and Challenges on the AI-Enabled Battlefield"
9:00am - PKI 158 - https://unl.zoom.us/j/94268094742 - Juan Lopez Jr, Dept. of Homeland Security, "Opportunities and Challenges in Critical Infrastructure Security Research"
2:00pm - PKI 164 - https://unl.zoom.us/j/98058853637
01 - Smart Health
10:30am-12:00pm - PKI 161 - https://unl.zoom.us/j/94107379509
- SH-01: "AIoT for Wireless Health", Honggang Wang, Yeshiva University
- SH-02: "Enhancing Privacy in Healthcare: Integrating Federated Learning and Differential Privacy for Sensitive Health Data", Dongfeng Fang, California Polytechnic State University
- SH-03: "Hybrid Federated Learning for E-Health with Horizontal and Vertical Data Partitioning", Chong Yu, University of Cincinnati
02 - AI and Wireless Networks
10:30am-12:00pm - PKI 250 - https://unl.zoom.us/j/91949122307
- AI-01: "From Uberized Wireless Multiple Access To Semantic Multimedia Communications", Wei Wang, San Diego State University
- AI-02: "Securing Machine Learning Systems in an Evolving Threat Landscape", Shengjie Xu, University of Arizona
- AI-03: "User Location Prediction for IRS-Assisted Communication Systems in Mobile Environments", Pejman Ghasemzadeh, Oklahoma State University
- AI-04: "NLP and Generative AI", Kimia Ameri, Eli Lilly and Company
03 - Internet of Things
10:30am-12:00pm - PKI 313 - https://unl.zoom.us/j/98221521538
- IOT-01: "Real-time Subsurface Sensing and Mapping with Cognitive Networked Robotic System", Dalei Wu, University of Tennessee at Chattanooga
- IOT-02: "NXP Connectivity and Security, An Overview", Subharthi Banerjee, NXP Semiconductors
- IOT-03: "Comprehensive Analysis of Diverse IoT Intrusion Data via Explainable Deep Learning Techniques", Sohan Gyawali, East Carolina University
- IOT-04: "Causal Inference-based Feature Selection and Clustering for Battery Pack Manufacturing", Shuaiqi Shen, University of Wisconsin-Milwaukee
04 - Wireless Network Management
10:30am-12:00pm - PKI 108 - https://unl.zoom.us/j/94327914090
- WNM-01: "Game Theory-based Resource Allocation for Vehicular Communication Networks", Kun Hua, California Polytechnic State University
- WNM-02: "AI-Supported Networking Analytics and Next-Generation Wireless Systems", Feng Ye, University of Wisconsin-Madison
- WNM-03: "BatchMDS: A Time-efficient Misbehavior Detection System in Internet of Vehicles", Yili Jiang, Georgia State University
- WNM-04: "Semi-supervised Federated Learning for Misbehavior Detection of BSMs in Vehicular Networks", Jiaqi Huang, University of Central Missouri
Keynotes
"Rise of the Machines? Opportunities and Challenges on the AI-Enabled Battlefield"
James M. Taylor, Jr.
Nebraska Defense Research Corporation
Abstract:
The advent of more powerful and sophisticated intelligent systems has led to the deployment of operational systems with remarkable autonomous capabilities across many fields, including transportation, space exploration, manufacturing, agriculture, healthcare, environmental monitoring, cybersecurity, and even defense. These intelligent systems have improved the efficiency, effectiveness, and precision of tasks previously performed by humans by moving more of the repetitive, high-accuracy, or time-critical decisions out of the control of a human pilot, driver, or machine operator and into a domain governed by an algorithm that can learn from the environment, make an assessment of the situation, decide on a course of action, and execute that course of action much faster than any human could. As a result of this type of decision automation, humans have sent spacecraft billions of miles from Earth to conduct reconnaissance on planets and their moons, built self-driving cars that safely move passengers from place to place, and fully autonomous robotic surgical systems that could operate on a patient with minimal human input.
When it comes to defense-related systems, however, the fact that we can deploy an intelligent system or AI-enabled capability to the battlefield doesn’t necessarily mean that we should deploy that capability -- even if it gives our side a military advantage over the adversary. By their very nature, military systems designed to inflict lethal force must also be used in accordance with the law of war, applicable treaties, and humanitarian law.
For this talk, we will examine recent advancements in the decision-making capabilities of intelligent systems on the battlefield using Colonel John Boyd’s Observe-Orient-Decide-Act decision process as a frame of reference. In addition, we will explore the challenges of leveraging artificial intelligence capabilities for decision support in lethal systems, given difficulties in data collection, data cleaning, algorithm training, and achieving a sufficiently low error rate. Finally, we will consider the legal and moral implications of deploying lethal intelligent systems on the battlefield and recommend areas for further research.
Bio:
Dr. James M. Taylor, Jr. (Jim) serves as Deputy Executive Director of the Nebraska Defense Research Corporation (NDRC), a non-profit affiliate of the University of Nebraska.
In this role, Jim supports NDRC’s mission to provide subject matter expertise, capability demonstrations, and market research in the Command, Control, and Communications (C3) and other mission domains for our United States Strategic Command sponsor.
Prior to joining NDRC in 2021, Jim served as the Director for STRATCOM Mission Systems of the National Strategic Research Institute (NSRI) and as a Principal Investigator or Co-Investigator on research projects involving consequence management, decision support, information technology and operational technology cybersecurity, and civilian leadership development. Additionally, Jim spearheaded the establishment of NSRI’s Academic Wargaming Center, in coordination with University of Nebraska researchers, and developed innovative wargaming tools to build capacity in decision making, critical thinking, and creative problem solving.
Jim brings 20 years of active duty U.S. Air Force developmental engineering and acquisition experience to the NDRC team, having retired in 2012 at the rank of Lieutenant Colonel. Dr. Taylor holds a Bachelor of Electrical Engineering from Georgia Tech, a Master of Science in Electrical Engineering from the Air Force Institute of Technology, and a Doctorate of Computer Engineering from the University of Nebraska - Lincoln.
"Opportunities and Challenges in Critical Infrastructure Security Research"
Juan Lopez Jr
Department of Homeland Security
Abstract:
Critical Infrastructure Security Research is a complex domain. Current approaches are useful in solving fundamental research problems in this area and add value to the scientific body of knowledge. However, in practice, they rarely adequately address the full range of challenges faced in DHS operational environments. Preparing the next generation of problem solvers to achieve success in this domain warrants exploring approaches to enhance their skill set in real-time while progressing through their academic journey. I will discuss the value of developing multidisciplinary thinking and opportunities to apply these skills during DHS internships.
Bio:
Juan Lopez Jr., Ph.D., CISSP, is currently a Lead Scientific Consultant in Industrial Control Systems and Open-Source Software under the Critical Infrastructure Security and Resilience Research (CISSR) program at DHS Science & Technology Directorate. His prior work at Oak Ridge National Laboratory included serving as Group Leader, Energy and Control Systems Security, National Security Sciences Directorate.
He leads research in Critical Infrastructure Protection, Supervisory Control and Data Acquisition (SCADA) systems, Nuclear Power Cybersecurity, Electromagnetic Interference (EMI) modeling, and UWB Communications. He led research for DOE under the Cybersecurity for Energy Delivery Systems (CEDS), DOD sponsors, and provided cybersecurity training for the Defense Threat Reduction Agency (DTRA). He holds a patent in Radio Frequency-DNA fingerprinting.
He served as the technical lead in SCADA/ICS research at the Air Force Cyberspace Technical Center of Excellence located at the Air Force Institute of Technology on Wright-Patterson Air Force Base, Ohio. Lopez earned a PhD in Computer Science at the Air Force Institute of Technology, Bachelor of Science from the University of Maryland, Master of Science in Information & Telecommunications Systems Management from Capitol College, and a Master of Science from the Air Force Institute of Technology under the NSA’s Information Assurance Scholarship Program. Lopez is an IEEE Senior Member, ACM member, Co-Chair for the Industrial Society of Automation’s Work Group 4, Task Group 7 (Security of ICS Sensors), Certified Information Systems Security Professional (CISSP), Certified SCADA Security Architect (CSSA), Certified Scrum Master, Lean Six Sigma Green Belt, and has an Extra Class amateur radio license from the Federal Communication Commission (FCC). He served on active duty for 27 years in the U.S. Marine Corps and is a distinguished pistol shot.
Track 1 - Smart Health
SH-01: "AIoT for Wireless Health"
Honggang Wang
Yeshiva University
Abstract:
Artificial Intelligence of Things (AIoT) for Wireless health is becoming a popular research area where the Internet, sensing, wireless communications, and intelligent techniques (e.g., AI and data analytics) are used to support digital health applications. AIoT systems, such as smart wearable systems with various types of biomedical sensors, are the critical infrastructures of wireless health and provide an opportunity to address issues in rapidly increasing digital health applications. However, there are significant challenges in this area, such as improving the performance of wearables, analyzing large and continuous physiological data streams collected from biomedical sensors, building predictive models, and data transmission, especially in mobile and wireless environments. In this lecture, I will introduce our related studies.
Bio:
Honggang Wang is a professor and Chair of Graduate Computer Science and Engineering Department at Katz School of Science and Health, Yeshiva University. He was a professor of Electrical and Computer Engineering at UMass Dartmouth. He received his Ph.D. degree from University of Nebraska-Lincoln in 2009. His recent research interests include Artificial Intelligence and its applications to digital health and autonomous vehicles, 5G/6G communications, and cybersecurity. He has produced high-quality publications in prestigious journals and conferences in his research areas, winning six prestigious best paper awards. He is an alumnus of the National Academic Engineering (NAE) Frontiers of Engineering program. He serves as the steering committee and founding co-chair of the IEEE/ACM Conference on Connected Health (CHASE), a leading international conference in connected health. He served as the Editor in Chief (EiC) of IEEE Internet of Things Journal (2020 -2022). He was the past Chair (2018-2020) of IEEE Multimedia Communications Technical Committee and IEEE eHealth Technical Committee (2020-2021). He was named an IEEE Fellow for his contribution to IoT and multimedia applications. He is a fellow of Asia-Pacific Artificial Intelligence Association (AAIA).
SH-02: "Enhancing Privacy in Healthcare: Integrating Federated Learning and Differential Privacy for Sensitive Health Data"
Dongfeng Fang
California Polytechnic State University
Abstract:
The widespread collection and analysis of sensitive health data has raised significant concerns about privacy and security in healthcare applications. Federated learning (FL) has emerged as a promising solution to these challenges by enabling machine learning models to be trained across distributed systems without sharing the original sensitive health data. However, despite its potential, FL alone does not fully address the risk of privacy breaches, particularly from inference attacks. To further safeguard sensitive health data, differential privacy (DP) can be integrated into the FL framework. This study explores the application of FL combined with DP on sensitive health datasets, aiming to enhance privacy-preserving capabilities while maintaining robust model performance. We review the fundamental principles of FL and DP, analyze their integration, and evaluate the trade-offs between analysis accuracy and privacy in healthcare applications.
SH-03: "Hybrid Federated Learning for E-Health with Horizontal and Vertical Data Partitioning"
Chong Yu
University of Cincinnati
Abstract:
E-Health systems generate large volumes of sensitive data from various sources, posing challenges in privacy, communication, and computational efficiency. Federated Learning (FL) addresses these challenges by enabling edge devices to train models collaboratively while keeping data private. However, the diverse and heterogeneous nature of medical data limits the efficiency of traditional FL methods, and simple combinations of Horizontal and Vertical FL often result in poor training performance and unstable convergence. This talk introduces a Hybrid Federated Learning framework with a three-tier architecture to efficiently manage diverse data. We propose a Hybrid Stochastic Gradient Descent algorithm, which ensures secure, accurate, and fast training. Through convergence analysis and adaptive communication strategies, the framework provides a scalable and efficient solution tailored to the needs of e-health systems.
Track 2 - AI and Wireless Networks
AI-01: "From Uberized Wireless Multiple Access To Semantic Multimedia Communications"
Wei Wang
San Diego State University
Abstract:
In this talk, the concept of uberizing wireless access will be presented, to address the explosive growth of the number of mobile users in future generations of wireless multiple access networks. The medium access control problem of wireless users will be modeled using the uber/lyft driver-passenger shared economics concept. Then a game-theoretic solution with a two-stage Stackelberg game will be discussed and the Nash Equilibrium will be derived using a backward induction method. An incentive-driven content shifting framework will be presented to reward users which push multimedia data content from an Orthogonal Multiple Access (OMA) link to a Non-Orthogonal Multiple Access (NOMA). In addition, a cryptocurrency concept named NOMAToken will be discussed to strategically reward the altruistic cache and relay behaviors with smart contracts among local wireless neighbors. Finally, ongoing research of semantic multimedia communications and machine learning-based multimedia compression-communication will be discussed.
Bio:
Dr. Wei Wang is a Professor and Graduate Advisor of Computer Science at San Diego State University. He joined San Diego State University in August 2014. He was an Assistant Professor of Computer Science at South Dakota State University from January 2010 to May 2014. He received his Ph.D. degree in Computer Engineering from the University of Nebraska - Lincoln in December 2009. His research interests include wireless networks, semantic multimedia communications, QoE-QoS issues, network economics and blockchains. He serves as the symposium co-chair of ICNC-MCC 2024, the publicity co-chair of IEEE/ACM IWQoS 2021, the publication co-chair of IEEE INFOCOM 2020, the area chair of IEEE ICME 2019-2020, the symposium co-chair of IEEE GLOBECOM-NGNI 2019, the web co-chair of IEEE INFOCOM 2016-2018, the symposium co-chair of IEEE ICC-NGNI 2018, and the vice chair of IEEE MMTC 2022-2024.
AI-02: "Securing Machine Learning Systems in an Evolving Threat Landscape"
Shengjie Xu
University of Arizona
Abstract:
As artificial intelligence (AI) and machine learning (ML) become integral to modern technologies, the security of these systems has never been more critical. This talk will explore the unique security challenges posed by AI/ML systems and offer practical strategies to mitigate associated risks. Using a framework that integrates security throughout the ML pipeline, we will examine how the MLOps lifecycle intersects with security best practices, identifying key opportunities for securing models at every stage.
In addition, I will introduce the Arizona Cybersecurity Clinic, a Google-funded initiative where I serve as co-PI. This clinic provides free digital security services to organizations, engaging faculty and students in conducting vulnerability assessments, security audits, compliance checks, and policy development. Modeled after community clinics in law or medicine, this initiative enhances local cybersecurity resilience while preparing students to be highly skilled defenders in the digital age. The clinic also plays a pivotal role in advancing our community’s understanding of cyber threats and fostering the development of innovative cybersecurity solutions.
AI-03: "User Location Prediction for IRS-Assisted Communication Systems in Mobile Environments"
Pejman Ghasemzadeh
Oklahoma State University
Abstract:
Intelligent Reflecting Surfaces (IRS) have emerged as one of the most promising technologies in modern wireless communication for addressing the unique channel effects in the super and extremely high-frequency spectrum. However, several challenges must be accounted for before IRS can be practically implemented. Particularly, two critical challenges in IRS-assisted communication in mobile environments are: (1) delays in user localization especially considering mmWave communication, and (2) delays in adjusting the IRS elements’ phases by access point/base station in wireless settings. The accumulation of these delays prevents the optimal IRS elements’ phase adjustment in real-time. In our current research, we investigate the importance of real-time user location prediction to ensure optimal phase adjustment of IRS elements in order to maintain communication performance. To accomplish this, we study three velocity regions (high, intermediate and low) in rectilinear and Brownian motion scenarios in which the impact of user location prediction on communication performance is demonstrated. Our future research will focus on exploring the role of user location prediction in more complex, real-world IRS-assisted communication environments.
AI-04: "NLP and Generative AI"
Kimia Ameri
Eli Lilly and Company
Bio:
Kimia Ameri is an Advisor and Data Scientist specializing in Natural Language Processing (NLP) within the Advanced Analytics and Data Sciences division at Eli Lilly and Company. With deep expertise in Generative AI, she focuses on fine-tuning large language models and integrating advanced techniques like LoRA and QLora to optimize fine-tuned LLMs for document generation. Her work also includes innovative projects such as Agent-Based Authoring Refinement which have been pivotal in refining AI-driven content generation.
Kimia's work bridges cutting-edge research and practical applications, driving significant improvements in efficiency and accuracy within regulated environments. Her contributions have been instrumental in enhancing large-scale enterprise systems, particularly in the medical and pharmaceutical sectors.
Track 3 - Internet of Things (IoT)
IOT-01: "Real-time Subsurface Sensing and Mapping with Cognitive Networked Robotic System"
Dalei Wu
University of Tennessee at Chattanooga
Abstract:
Underground utilities are vital infrastructure for modern cities. However, much of this infrastructure is aging or exists in unknown conditions and locations. Effectively managing, maintaining, and constructing subsurface utilities requires advanced monitoring methods and tools. Unfortunately, current subsurface sensing technologies fall short in terms of timeliness, versatility, intelligence, accuracy, and human-machine interfaces. The precise detection, localization, and assessment of underground infrastructure remain significant technical challenges, necessitating innovative solutions. This presentation will introduce a cutting-edge cyber-physical approach to monitoring and mapping underground utilities, integrating advanced technologies such as sensing, networking, control, edge computing, and machine learning. The project is funded by the NSF Cyber-Physical Systems (CPS) program.
Bio:
Dr. Dalei Wu earned his Ph.D. in Computer Engineering from the University of Nebraska-Lincoln in December 2010. From 2011 to 2014, he worked as a Postdoctoral Researcher with the Mechatronics Research Laboratory at Massachusetts Institute of Technology (MIT). In August 2014, he joined the Department of Computer Science and Engineering at the University of Tennessee at Chattanooga as a tenure-track assistant professor. He now holds the titles of Guerry Professor and UC Foundation Professor. Dr. Wu's research centers on cyber-physical systems and data-driven intelligent systems. His research is supported by the National Science Foundation (NSF) and the State of Tennessee.
IOT-02: "NXP Connectivity and Security, An Overview"
Subharthi Banerjee
NXP Semiconductors
Abstract:
Our digitally enhanced world is evolving to anticipate and automate. NXP Semiconductors N.V. enables a smarter, safer and more sustainable world through innovation. As the world leader in secure connectivity solutions for embedded applications, NXP is pushing boundaries in the automotive, industrial & IoT, mobile, and communication infrastructure markets. With cutting edge technologies in Wi-Fi, Bluetooth, Ultra-Wide-band (UWB) and 5G, NXP's ever growing portfolio supports industries demand in high-data rate, low-latency, energy-efficient wireless communication. This presentation will explore some of the innovations, applications and trends in connectivity that enable wireless researchers and engineers to address real-world challenges in connectivity and communication.
IOT-03: "Comprehensive Analysis of Diverse IoT Intrusion Data via Explainable Deep Learning Techniques"
Sohan Gyawali
East Carolina University
Abstract:
The proliferation of diverse Internet of Things (IoT) devices has significantly enhanced our daily lives, improving efficiency and service quality. However, this increase in the number of IoT devices has been accompanied by a corresponding escalation in cyber-attacks directed at these devices, posing potential threats to systems reliant on these devices. To address these issues, my main objective is to generate an extensive range of IoT intrusion data and subject it to analysis using explainable deep learning approaches. The ultimate objective is to create a mechanism capable of interpreting the outcomes of deep learning-driven detection and applying them to automate intrusion response operations for IoT systems. The research plan encompasses several key components: (1) Formulating various cyber-attack models for IoT systems, (2) Generating real-time datasets within a dedicated IoT system environment encompassing various cyber-attack scenarios, (3) Conducting an exhaustive evaluation of distinct explainable deep learning techniques for the identification and pinpointing of cyber-attacks, (4) Devising an innovative automated intrusion response system that takes an action based on outcomes from explainable deep learning techniques.
IOT-04: "Causal Inference-based Feature Selection and Clustering for Battery Pack Manufacturing"
Shuaiqi Shen
University of Wisconsin-Milwaukee
Abstract:
Understanding the causal relationships between features and labels is critical in building robust and interpretable machine learning models, particularly in complex manufacturing processes. This research investigates the impacts of battery cell features measured during manufacturing processes on inconsistency among cells to improve cell grouping for battery pack formation and minimize test failure rate. Specifically, we propose a novel feature selection approach by employing causal AI techniques to identify and establish the most significant factors that directly influence the voltage deviation of battery packs. In addition, a clustering method is developed based on these causally crucial features to enhance model performance and interpretability. Furthermore, by leveraging a real-world dataset provided by our industrial collaborator, we evaluate the proposed causal inference-based approach in clustering accuracy compared with the existing correlation-based techniques. The proposed approach also provides insights into the underlying relationships between process variables and final product quality. Our findings highlight the potential of combining causal inference with predictive AI modeling to develop interpretable, data-driven solutions in manufacturing and other industrial applications.
Track 4 - Wireless Network Management
WNM-01: "Game Theory-based Resource Allocation for Vehicular Communication Networks"
Kun Hua
California Polytechnic State University
Abstract:
To meet the low latency demands of emerging Communication, Control, and Computation (CCC) intensive vehicular networks, offloading these services to the roadside units, edge/cloud servers, and Unmanned Aerial Vehicles (UAV) has proven to be an effective approach. However, the limited network resources present significant challenges in developing an efficient resource allocation strategy. Balancing latency, throughput, and resource utilization is considered to be critical to ensure optimal performance in such ad hoc communication networks. In this study, we investigated the throughput and transmission delay performances for real-time and delay-sensitive services using a repeated game-theoretic approach. By employing Nash Equilibrium within a non-cooperative game model, we analyzed the efficiency of resource allocation strategies. Simulation results demonstrated notable improvements in resource allocation efficiency, validating the effectiveness of our approach.
WNM-02: "AI-Supported Networking Analytics and Next-Generation Wireless Systems"
Feng Ye
University of Wisconsin-Madison
Abstract:
Future networking and wireless systems require efficient network traffic analytics, seamless connectivity for Internet-of-Things (IoT) devices, and low-latency wireless communications, to support the emergence of new applications such as digital twin, e-business and smart manufacturing, as well as skyrocketing penetration of massive network devices. This talk will first present a new paradigm of sustainable AI-supported network traffic analytics, including self-evaluation on performance and self-maintained knowledge database for a network traffic classifier (NTC), acceleration on NTC implementation, resource-aware network quality-of-service clustering, and a simulation platform. The second part of this talk will present an IoT communication platform with the capabilities of autonomous technology detection, seamless transmit power control, and seamless IoT security. The third part of this talk will present fast channel state information (CSI) processing in the next-generation massive multiple-input-multiple-output communication systems. The focus in this part will be on closing the gap between the current AI-based CSI processing and the coherence time requirement in massive MIMO systems.
WNM-03: "BatchMDS: A Time-efficient Misbehavior Detection System in Internet of Vehicles"
Yili Jiang
Georgia State University
Abstract:
The Internet-of-Vehicles (IoV) has gained significant attention from both academia and industry, driven by its potential to enhance road safety and traffic efficiency. To fully realize this potential, securing IoV communications is essential, where detecting malicious Basic Safety Messages (BSMs) is a key challenge. To this end, machine learning (ML) has been widely employed to design Misbehavior Detection Systems (MDSs) for identifying manipulated BSMs. However, the existing MDSs mainly focus on improving detection accuracy, with little attention given to time efficiency. As safety in IoV is time-sensitive, designing a time-efficient MDS is urgently necessary.
In this talk, Dr. Jiang will introduce BatchMDS, a time-efficient MDS using Convolutional Neural Network (CNN) and data-to-image transformation. The rationale behind BatchMDS is allowing to read a batch of BSMs in one image, significantly reducing detection time compared with reading BSM one by one in the state-of-the-art. The experimental results demonstrate that BatchMDS can improve time efficiency by up to 84% while remaining remarkable detection accuracy.
WNM-04: "Semi-supervised Federated Learning for Misbehavior Detection of BSMs in Vehicular Networks"
Jiaqi Huang
University of Central Missouri
Abstract:
Basic Safety Messages (BSMs) exchanged among vehicles and roadside units through vehicular communications can significantly enhance road safety and improve traffic efficiency. Protecting the integrity of BSMs, which are transmitted wirelessly in plaintext, is critical for the proper operation of vehicular networks. As a result, various machine learning-based misbehavior detection systems have been proposed to identify corrupted BSMs. Recent studies have applied federated learning methods to further preserve user privacy while facilitating detection model updates. However, supervised federated learning cannot be directly applied since BSMs received by vehicles are unlabeled. In this paper, we propose a semi-supervised federated learning framework that enables the federated training process between vehicles and the server without transmitting any datasets. Our experimental results show that the misbehavior detection performance of the proposed semi-supervised framework is very close to the centralized method, while preserving user privacy and reducing communication costs.