Over the years, there has been an increase in the use of wearable sensors for high-precision Human Activity Recognition (HAR), ranging from personal fitness to remote patient monitoring. However, the continuous collection of biometric information poses a privacy risk, potentially leading to user profiling and data misuse. Today’s state-of-the-art approaches use standard encryption techniques to protect data in transit but leave it vulnerable during computation. For accurate HAR while maintaining the privacy of users’ data during collaborative analysis, we propose a hybrid, domain-agnostic framework that integrates Homomorphic Encryption (HE) with Secure Multi-Party Computation (SMPC). The proposed approach enables healthcare providers and device manufacturers to collaboratively analyze data without ever disclosing raw biometrics by utilizing HE for data encryption and secret sharing for collaborative computation. While this framework applies to general HAR scenarios, it will be crucially useful in the high-stakes domain of eldercare, where data privacy and regulatory compliance are critical. Reliable, Privacy-Aware Human Activity Sensing (RAHAS) achieved \(89.24\pm 0.95\%\) accuracy when tested on the widely used PAMAP2 dataset for HAR. Our analysis demonstrates that our privacy-preserving design provides side-channel resilience while maintaining utility comparable to clear-text baselines.

This paper presents HyFuzz, a hybrid intelligent fuzz testing platform designed to enhance the security validation of next generation (NextG) mobile networks. HyFuzz integrates symbolic formal analysis with adaptive fuzzing to enable the discovery of vulnerabilities that emerge from subtle state inconsistencies and session level command manipulations. Specifically, HyFuzz demonstrates support for multi step intra session fuzzing, where carefully crafted command sequences cause persistent state desynchronization between User Equipment (UE) and the network. Complementing this, HyFuzz employs formal guided deep fuzzing, directing fuzzing efforts to high risk protocol states identified by symbolic analysis. Through a dual mode architecture supporting both virtual (ZMQ) and over the air (OTA) fuzzing, HyFuzz provides an extensible testbed for low level and behavioral vulnerability discovery. Experimental results across 1,281 test cases reveal 1,105 failure instances, including stealthy failures that manifest only under extended interaction contexts. Our findings suggest HyFuzz provides a foundational capability toward more realistic and semantically rich vulnerability detection in modern mobile infrastructure.

Public health agencies face critical challenges in identifying high-risk neighborhoods for childhood lead exposure with limited resources for outreach and intervention programs. To address this, we develop a Priority Score integrating untested children proportions, elevated blood lead prevalence, and public health coverage patterns to support optimized resource allocation decisions across 136 neighborhoods in Chicago, New York City, and Washington, D.C. We leverage these allocation tasks, which require integrating multiple vulnerability indicators and interpreting empirical evidence, to evaluate whether large language models (LLMs) with agentic reasoning and deep research capabilities can effectively allocate public health resources when presented with structured allocation scenarios. LLMs were tasked with distributing 1,000 test kits within each city based on neighborhood vulnerability indicators. Results reveal significant limitations: LLMs frequently overlooked neighborhoods with highest lead prevalence and largest proportions of untested children, such as West Englewood in Chicago, while allocating disproportionate resources to lower-priority areas like Hunts Point in New York City. Overall accuracy averaged 0.46, reaching a maximum of 0.66 with ChatGPT 5 Deep Research. Despite their marketed deep research capabilities, LLMs struggled with fundamental limitations in information retrieval and evidence-based reasoning, frequently citing outdated data and allowing non-empirical narratives about neighborhood conditions to override quantitative vulnerability indicators.

Quantum Neural Networks (QNNs) offer promising capabilities for complex data tasks, but are often constrained by limited qubit resources and high entanglement, which can hinder scalability and efficiency. In this paper, we introduce Adaptive Threshold Pruning (ATP), an encoding method that reduces entanglement and optimizes data complexity for efficient computations in QNNs. ATP dynamically prunes non-essential features in the data based on adaptive thresholds, effectively reducing quantum circuit requirements while preserving high performance. Extensive experiments across multiple datasets demonstrate that ATP reduces entanglement entropy and improves adversarial robustness when combined with adversarial training methods like FGSM. Our results highlight ATP’s ability to balance computational efficiency and model resilience, achieving significant performance improvements with fewer resources, which will help make QNNs more feasible in practical, resource-constrained settings.

The freshness of information related to status updates is crucial in time-critical applications like disaster response and search and rescue operations. We can enhance network connectivity by using electric vertical take-off and landing vehicles (eVTOLs) operating in the affected region as portable wireless repeaters as part of the operation. In this work, we study the spectrum allocation in ultra-reliable low latency communication (URLLC) networks assisted by eVTOLs while minimizing the age of information (AoI). The optimal spectrum allocation for eVTOL-assisted networks is a challenging problem that depends on various dynamic factors, such as bit error rate (BER), data rate, power consumption, and the flight trajectory of the eVTOLs. Therefore, we propose a dynamic approach that can efficiently allocate spectrum among the eVTOLs to minimize the total AoI between the source and the destination, as well as the individual AoI of each eVTOL acting as a relay node. Simulations exhibit that the proposed algorithm can outperform the classical approaches in terms of AoI by improving the AoI by $57.42 \%$ over the classical fairness approach and $43.43 \%$ over the classical optimal relay selection approach. We also find an improvement in data rate by $8.89 \%$ and $6.76 \%$, respectively, along with a marginal improvement in the BER. Our approach offers an efficient solution for next-generation AoI-aware spectrum management in eVTOL-assisted networks.

The reliability of 5G systems and their applications in a complex, dynamic, and heterogeneous environment requires rigorous testing and real-time detection for system vulnerabilities and unintended emergent behaviors. In this paper, fuzz testing is performed on 5G systems by randomly injecting and permuting control commands into the system under test (SUT) of the 5G radio resource control (RRC) authentication and authorization process, emulating Man-In-The-Middle (MITM) attacks to trigger potential vulnerabilities and unintended behaviors. The fuzzed system behaviors contain information that could indicate the system's health status, and potential vulnerabilities, and, more importantly, it enables the causation analysis in the SUT to detect the location and type of attacks or abnormal inputs from the profiling of the impacted behaviors. We then propose a Real-time Framework for Root Cause Analyses (RAFT) in NextG Vulnerability Detection based on analyzing the random fragments of the log file generated during the communication process. By processing the random fragments of the logging profiles captured during fuzz testing with the continuous bag-of-words (CBOW) Model, we extract the information of states and states transitions and perform causal analysis to identify the root cause for vulnerability detection in the 5G system. The novelty of our framework lies in the creation and analysis of the information extraction that does not require capturing the entire log file instead only the log file fragments to achieve high accuracy. This approach enables real-time detection and deployment to real-life scenarios where access to the entire logging profile is difficult to obtain or unavailable. The presented framework RAFT directly adapts to various machine learning (ML) models, which allow the adaptation to hardware with various computation complexity from internet-of-things (IoT) to Radio Access Network (RAN) servers. The experimental results show a significant performance gain and are thoroughly evaluated by the accuracy and area under the curve (AUC) results. In particular, we show that the proposed framework can attain a high AUC value ($0.92\leq\text{AUC} < 0.96$) by accessing only a 70% fragment of the original log file while maintaining a higher accuracy. In addition, we find that RAFT reduces the time complexity by more than 5% as the fragment size reduces to 70% of the original log file. The causation analysis nature of RAFT summarizes vulnerability information into essential root causes that can be easily transmitted within the network in real-time and turned into guidance for back-end engineers. The unique advantages of RAFT, including accurate causation with information fragments, reliable performance without large training datasets and less computation complexity guarantee a wide range of use cases and deployment environment of RAFT.

The 5G authentication and key agreement protocol (AKA) is intended to provide security assurance for users, net-work operators, and machine-to-machine communications. It is a critical component of everyday life and national infrastructure. In this paper, we conduct a high-resolution and in-depth formal analysis of the AKA protocol, addressing hidden assumptions stemming from the underlying dependencies among entities. We categorize the direct properties among identifiers in each protocol session into confidentiality, integrity, authentication, and accounting. We further uncover the indirect dependencies among identifiers through the propagation of assumptions in the designed dependency graph. The formal models are generated from both the direct and indirect dependencies and are verified using Pro Verif. Our approach reveals four major vulnerability types across four sub-procedures. This includes three vulner-abilities previously identified in existing research, as well as one newly discovered vulnerability type. The solutions proposed to address these vulnerabilities have been validated through formal verification and over-the-air (OTA) testing. We present the first formal models that consider hidden assumptions and their propagation in 5G, and demonstrate the fragility of the 5G-AKA protocol through experimental practice. We have also included formally verified fixes for the encountered vulnerabilities.

Controlling the carbon footprint and operating costs of 5G and nextG networks remains a venerable problem among network designers aiming for high spectrum efficiency and communication performance. This paper introduces a Game-Theoretic solution known as GeTOA (Game-Theoretic Age of Information), which leverages Nash bargaining (NB) to optimize the Age of Information (AOI) for multi-user electric vertical take-off and landing (eVTOL) communication. Considering the unique trajectory of the e VTOLs, which have substantial alterations in the channel conditions, coupled with the variation in the AOI during critical phases of flight, we calculate the Pareto optimal solutions for fair and efficient use of available resources while increasing the information content in our communication. We compare the cooperative GeTOA approach against the non-cooperative utility maximization (UM) approach, resulting in a notable 12.76% improvement while ensuring equitable resource allocation among the nodes. In contrast to the traditional UM approach, GeTOA significantly enhances energy allocation efficiency for multiple e VTOLs operating with diverse trajec-tories while enabling zero-touch fair resource management in open-access spectrum scenarios. In particular, results show that GeTOA handles fairness among the eVTOLs by ensuring an equal rate of information and fair distribution of power at the eVTOLs, which is especially relevant for Citizens Broadband Radio Service (CBRS) and C- Band applications. The flexibility and prioritization of AOI-based optimization allow a significant number of e VTOLs to operate efficiently within a congested spec-trum, facilitating Ultra-Reliable Low Latency Communication (URLLC) and improving power efficiency for the advancement of large-scale Urban Air Mobility (UAM) services which the limited flight range of e VTOLs has historically restricted.

Despite the promises of current and future cellular networks to increase security, privacy, and robustness, 5G networks are designed to streamline discovery and initiate connections with limited computation and communication costs, leading to the predictability of control channels. This predictability enables signal-level attacks, particularly on unprotected initial access signals. To assess vulnerability in access control and enhance robustness in cellular networks, we present a strategic approach leveraging O-RAN architecture in this paper that detects and classifies signal-level attacks for actionable countermeasure defense. We evaluate attack scenarios of various power levels on both 4G/LTE-Advanced and 5G communication systems. We categorize the types of attack models based on the attack cost: Overshadowing and Jamming. Overshadowing represents low attack power categories with time and frequency synchronization, while Jamming represents un-targeted attacks that cause similar quality-of-service degradation as overshadowing attacks but require high power levels. Our detection strategy relies on supervised machine-learning models, specifically a Reservoir Computing (RC) based supervised learning approach that leverages physical and MAC-layer information for attack detection and classification. We demonstrate the efficacy of our detection strategy through extensive experimental evaluations using the O-RAN platform with software-defined radios (SDRs) and commercial off-the-shelf (COTS) user equipment (UEs). Empirical results show that our method can classify the change in statistics caused by most overshadowing and jamming attacks with more than 95% classification accuracy.

The increased flexibility and density of spectrum access in 5G New Radio (NR) has made jamming detection and classification a critical research area. To detect coexisting jamming and subtle interference, we introduce a Bayesian Inference-assisted machine learning (ML) methodology. Our methodology uses cross-layer Key Performance Indicator data collected on a Non-Standalone (NSA) 5G NR testbed to leverage supervised learning models, further assessed, calibrated, and revealed using Bayesian Network Model (BNM)-based inference. The models can operate on both instantaneous and sequential time-series data samples, achieving an Area under Curve above 0.954 for instantaneous models and above 0.988 for sequential models including the echo state network (ESN) from the Reservoir Computing (RC) family, across various jamming scenarios. The 180 ms instantaneous detection time allows for continuous tracking of the dynamic jamming condition due to UE mobility. Our approach serves as a validation method and a resilience enhancement tool for ML-based jamming detection while also enabling root cause identification for observed performance degradation. The introduced BNM-based inference proof-of-concept is successful in addressing 72.2% of the erroneous predictions of the RC-based sequential detection model caused by insufficient training data samples, thereby demonstrating its near real-time applicability in 5G NR and Beyond-5G networks.

The paper proposes a protocol-independent Listen-and -Learn (LAL) based fuzzing system, which provides a systematic solution for vulnerabilities and unintended emergent behavior detection with sufficient automation and scalability, for 5G and nextG protocols and large-scale open programmable stacks. We use the relay model as our base and capture and interpret packets without prior knowledge of protocols imple-mentation. Radio Resource Control (RRC) is selected proof of concept of the proposed system. Our fuzzing architecture incorporates two abstractions of different dimension fuzzing-command-level and bit-level, and the proposed LAL fuzzing framework focuses on command-level fuzzing covering potential attacks by autonomously generating a comprehensive fuzzing case set. Our analysis of 39 RRC states successfully illustrates 129 vulnerabilities resulting in RRC connection establishment failure from 205 command-level fuzzing cases and reveals insights into exploitable vulnerabilities in each channel of RRC procedure. Furthermore, to assess risks and prevent potential vulnerability, we use the Long Short-Term Memory (LSTM) based model to perform a deep analysis of transaction states in sequenced commands. With the LSTM based model, we efficiently predict more than 95% connection failure at an average duration of 0.059 seconds after the fuzzing attack and provide sufficient time for proactive defense before RRC connection completion or failure, with an average of 3.49 seconds. The rapid vulnerability prediction capability also enables proactive defenses to potential attacks. The proposed fuzzing system offers sufficient automation, scalability, and usability to improve 5G security assurance, and could be used for existing and newly released protocols and stacks validation and real-time system vulnerability detection and prediction.

Efficient and precise detection of vulnerabilities in 5G protocols and implementations is crucial for ensuring the security of its application in critical infrastructures. However, with the rapid evolution of 5G standards and the trend towards softwarization and virtualization, this remains a challenge. In this paper, we present an automated Fuzz Testing Digital Twin Framework that facilitates systematic vulnerability detection and assessment of unintended emergent behavior, while allowing for efficient fuzzing path navigation. Our framework utilizes assembly-level fuzzing as an acceleration engine and is demonstrated on the flagship 5G software stack: srsRAN. The introduced digital twin solution enables the simulation, verification, and connection to 5G testing and attack models in real-world scenarios. By identifying and analyzing vulnerabilities on the digital twin platform, we significantly improve the security and resilience of 5G systems, mitigate the risks of zero-day vulnerabilities, and provide comprehensive testing environments for current and newly released 5G systems.

The dissemination of information integrity at unprecedented speed and scale is a new phenomenon with the potential for vast harm if used incorrectly, specially applied in healthcare and clinical data. Despite holding much promise, the usefulness for clinical research using data from wearable devices that record user’s health conditions is limited by its integrity pitfall. This study presents and demonstrates a detection framework to effectively identify integrity compromises of wearable data and map the compromises with user scenarios under environmental influence. Through the Bayesian Network Model (BNM), the framework performs causation analyses between use scenario and data impact and integrates auto-encoder based data impact anomaly detection and classification. The auto-encoder based data impact detection eliminate the requirement for pre-training data, and enables a real-time detection with average latency of 4.6s. The BNM based causal inference shows accurate inference of user scenario based on the data impact detection. The proposed framework will allow for back tracing the root causes of the integrity compromises and trigger real-time human intervention to improve system integrity. We demonstrated system performance through a simulated use case.

Jamming and intrusion detection are some of the most important research domains in 5G that aim to maintain use-case reliability, prevent degradation of user experience, and avoid severe infrastructure failure or denial of service in mission-critical applications. This paper introduces an anonymous jamming detection model for 5G and beyond based on critical signal parameters collected from the radio access and core network’s protocol stacks on a 5G testbed. The introduced system leverages both supervised and unsupervised learning to detect jamming with high-accuracy in real time, and allows for robust detection of unknown jamming types. Based on the given types of jamming, supervised instantaneous detection models reach an Area Under the Curve (AUC) within a range of 0.964 to 1 as compared to temporal-based long short-term memory (LSTM) models that reach AUC within a range of 0.923 to 1. The need for data annotation effort and the required knowledge of a vocabulary of known jamming limits the usage of the introduced supervised learning-based approach. To mitigate this issue, an unsupervised auto-encoder-based anomaly detection is also presented. The introduced unsupervised approach has an AUC of 0.987 with training samples collected without any jamming or interference and shows resistance to adversarial training samples within certain percentage. To retain transparency and allow domain knowledge injection, a Bayesian network model based causation analysis is further introduced.

In the process of developing software, code review is crucial. By identifying problems before they arise in production, it enhances the quality of the code. Finding the best reviewer for a code change, however, is extremely challenging especially in large scale, especially open source software stacks with cross functioning designs and collaborations among multiple developers and teams. Additionally, a review by someone who lacks knowledge and understanding of the code can result in high resource consumption and technical errors. The reviewers who have the specialty in both functioning (domain knowledge) and non-functioning areas of a commit are considered as the most qualified reviewer to look over any changes to the code. Quality attributes serve as the connection among the user requirements, delivered function description, software architecture and implementation through put the entire software stack cycle. In this study, we target on auto reviewer assignment in large scale software stacks and aim to build a self-learning, and self-correct platform for intelligently matching between a commit based on its quality attributes and the skills sets of reviewers. To achieve this, quality attributes are classified and abstracted from the commit messages and based on which, the commits are assigned to the reviewers with the capability in reviewing the target commits. We first designed machine learning schemes for abstracting quality attributes based on historical data from the OpenStack repository. Two models are built and trained for automating the classification of the commits based on their quality attributes using the manual labeling of commits and multi-class classifiers. We then positioned the reviewers based on their historical data and the quality attributes characteristics. Finally we selected the recommended reviewer based on the distance between a commit and candidate reviewers. In this paper, we demonstrate how the models can choose the best quality attributes and assign the code review to the most qualified reviewers. With a comparatively small training dataset, the models are able to achieve F-1 scores of 77% and 85.31%, respectively.