Cyber-Physical Systems (CPS) are increasingly leveraging Federated Learning (FL) to enable decentralized intelligence while preserving privacy across distributed devices. Federated adversarial learning (FAL) leverages FL and adversarial training to enhance model robustness against adversarial attacks while maintaining data privacy across decentralized, heterogeneous devices. While FAL strengthens CPS resilience against adversarial threats, variations in energy constraints, carbon emissions, computational capabilities, and latency requirements introduce additional complexity. These variations impact energy consumption, carbon emissions, and power source efficiency, creating a complex trade-off between sustainability and robustness. This underscores the critical need for standardized metrics to systematically evaluate and balance these competing factors in FAL-enabled CPS. In this paper, we propose three novel robustness metrics designed to quantify the interplay between energy efficiency, sustainability dimensions, and adversarial robustness in FAL setups for CPS. The proposed methodology accounts for diverse CPS scenarios, addressing factors such as emissions, energy consumption, latency, renewable energy, and low-energy devices with critical latency needs. We validate our approach through simulations in two setups, including a single-device environment to isolate device variability and a heterogeneous multi-device environment to evaluate architectural impacts. The results demonstrate the effectiveness of our proposed metrics in systemically quantifying the trade-off between sustainability and robustness in FAL-based CPS.

Recent studies have shown that sponge attacks can significantly increase the energy consumption and inference latency of deep neural networks (DNNs). However, prior work has focused primarily on computer vision and natural language processing tasks, overlooking the growing use of lightweight AI models in sensing-based applications on resource-constrained devices, such as those in Internet of Things (IoT) environments. These attacks pose serious threats of energy depletion and latency degradation in systems where limited battery capacity and real-time responsiveness are critical for reliable operation. This paper makes two key contributions. First, we present the first systematic exploration of energy-latency sponge attacks targeting sensing-based AI models. Using wearable sensing-based AI as a case study, we demonstrate that sponge attacks can substantially degrade performance by increasing energy consumption, leading to faster battery drain, and by prolonging inference latency. Second, to mitigate such attacks, we investigate model pruning, a widely adopted compression technique for resource-constrained AI, as a potential defense. Our experiments show that pruning-induced sparsity significantly improves model resilience against sponge poisoning. We also quantify the trade-offs between model efficiency and attack resilience, offering insights into the security implications of model compression in sensing-based AI systems deployed in IoT environments.

Vision-Language Models (VLMs) have achieved remarkable performance in various downstream tasks, such as image captioning and visual question answering (VQA), by learning joint representations from paired image-text data. However, the robustness of VLMs against adversarial attacks is a critical concern as these models become widely adopted in real-world applications. This study provides a comprehensive overview of the current state of research on the adversarial robustness of VLMs. We introduce a novel taxonomy that categorizes the diverse landscape of adversarial attacks on VLMs, considering factors such as the modality of the attack, the level of model access, and the attack objective. We explore various types of attacks, including jailbreak, backdoor, data poisoning, gradient-based, and patch-based attacks, and discuss their underlying principles and assumptions. Furthermore, we present a comparative analysis of existing defense strategies and their strengths and limitations. By identifying open challenges and future research directions, we aim to stimulate further exploration of this crucial area and contribute to the development of robust and trustworthy VLMs for real-world applications.

Stress is a major factor affecting mental health and long-term well-being, making its early detection essential for effective intervention. Wearable sensors have emerged as a promising solution for continuous stress monitoring, but current systems face two major limitations. First, they are single-dimensional, relying on either chest- or wrist-based sensors alone, which limits their ability to capture diverse physiological signals. Second, they typically require centralized data collection, raising significant privacy concerns since stress-related information is highly sensitive. This work introduces a distributed multimodal stress monitoring framework that addresses both challenges. The framework integrates synchronized signals from both chest and wrist sensors, providing a richer and more reliable representation of stress patterns. An attention-based fusion mechanism is introduced to adaptively weigh complementary information across modalities, while two deep learning architectures, BiLSTM and TS-BERT, are employed for temporal modeling. To ensure scalability and protect user privacy, we adopt Federated Learning (FL) for decentralized training across simulated edge devices. Experimental results demonstrate that TS-BERT achieves superior performance, reaching 99.93% test accuracy in FL, closely matching centralized outcomes.

The rapid growth of machine learning (ML) technologies has raised significant concerns about their environmental impact, particularly regarding energy consumption and carbon emissions. This comprehensive review examines the intersection of ML and sustainability, synthesizing research from 2014 to 2024 to provide a holistic view of sustainable ML practices. This systematic review, encompassing over 200 peer-reviewed publications, reveals a growing emphasis on quantifying and mitigating the environmental footprint of ML systems. Key findings include: (1) a 300% increase in sustainable ML research since 2020; (2) the emergence of specialized carbon footprint quantification tools for ML; and (3) promising advancements in energy-efficient algorithms and green computing infrastructure. This research identifies critical challenges, including the lack of standardized sustainability metrics and the need for more robust life-cycle assessments of ML systems. The review also highlights the potential of transfer learning, federated learning, and hardware innovations in reducing ML's environmental impact. The analysis culminates in a novel framework for implementing sustainable practices in ML projects and a detailed roadmap for future research. This work provides researchers, practitioners, and policymakers with crucial insights to drive the development of more environmentally responsible ML technologies, ultimately contributing to global sustainability goals.

With the proliferation of edge devices, there is a significant increase in attack surface on these devices. The decen-tralized deployment of threat intelligence on edge devices, coupled with adaptive machine learning techniques such as the in-context learning feature of Large Language Models (LLMs), represents a promising paradigm for enhancing cybersecurity on resource-constrained edge devices. This approach involves the deployment of lightweight machine learning models directly onto edge devices to analyze local data streams, such as network traffic and system logs, in real-time. Additionally, distributing computational tasks to an edge server reduces latency and improves responsiveness while also enhancing privacy by processing sensitive data locally. LLM servers can enable these edge servers to autonomously adapt to evolving threats and attack patterns, continuously updating their models to improve detection accuracy and reduce false positives. Furthermore, collaborative learning mechanisms facilitate peer-to-peer secure and trustworthy knowledge sharing among edge devices, enhancing the collective intelligence of the network and enabling dynamic threat mitigation measures such as device quarantine in response to detected anomalies. The scalability and flexibility of this approach make it well-suited for diverse and evolving network environments, as edge devices only send suspicious information such as network traffic and system log changes, offering a resilient and efficient solution to combat emerging cyber threats at the network edge. Thus, our proposed framework can improve edge computing security by providing better security in cyber threat detection and mitigation by isolating the edge devices from the network.

Federated Learning (FL) has recently experienced tremendous popularity due to its emphasis on user data privacy. However, the distributed computations of FL can result in constrained communication and drawn-out learning processes, necessitating the client-server communication cost optimization. The ratio of chosen clients and the quantity of local training passes are two hyperparameters that have a significant impact on the performance of FL. Due to different training preferences across various applications, it can be difficult for FL practitioners to manually select such hyperparameters. In this paper, we introduce FedAVO, a novel FL algorithm that enhances communication effectiveness by selecting the best hyperparameters leveraging the African Vulture Optimizer (AVO). Our research demonstrates that the communication costs associated with FL operations can be substantially reduced by adopting AVO for FL hyperparameter adjustment. Through extensive evaluations of FedAVO on benchmark datasets, we identify the optimal hyperparameters that are appropriately fitted for the benchmark datasets, eventually increasing global model accuracy by 6% in comparison to the state-of-the-art FL algorithms (such as FedAvg, FedProx, FedPSO). The code, data, and experiments have been made publicly available on our GitHub repository11https://github.com/speedlab-git/FedAVO.

The prevalence of road and highway accidents, largely attributed to driver distraction, highlights the critical need for an intelligent system that can assess driver alertness and provide timely alerts. Current solutions in the market are often characterized by their high costs and complex installation processes, which significantly limit their accessibility and practical application on a broader scale. In response to this challenge, our research introduces a cost-effective, real-time framework designed to monitor driver alertness utilizing the Raspberry Pi, a device known for its limited processing capabilities. This inherent limitation prompted us to implement several optimizations, which are elaborated upon within our study, to equip the Raspberry Pi with the ability to make real-time decisions effectively. Our proposed approach features a novel algorithm that integrates Haar Cascade and facial landmark detection techniques, enabling the rapid and precise identification of facial features, thereby surpassing the accuracy of existing leading methods. This system meticulously tracks facial points to evaluate driver attentiveness through indicators such as drowsiness, yawning, and unusual facial movements. It utilizes metrics including the Eye Aspect Ratio (EAR), Lips Movement Ratio (LMR), and Face Position Difference (FPD) to initiate driver alerts, thereby contributing to the prevention of potential accidents. Furthermore, our system is enhanced with a GSM module, facilitating emergency notifications to the vehicle owner in critical situations. Extensive testing of our framework, involving participants of diverse sizes, skin colors, and ages, has demonstrated its efficacy in sustaining driver awareness with minimal processing delays, even when deployed on devices with limited computational resources. This affirms the potential of our proposed solution to serve as a viable and scalable option for enhancing road safety through improved driver alertness monitoring.

Cyber-physical systems have become an essential part of the modern healthcare industry. The healthcare cyber-physical systems (HCPS) combine physical and cyber components to improve the healthcare industry. While H CPS has many advantages, it also has some drawbacks, such as a lengthy data entry process, a lack of real-time processing, and limited real-time patient visualization. To overcome these issues, this paper represents an innovative approach to integrating large language model (LLM) to enhance the efficiency of the healthcare system. By incorporating LLM at various layers, HCPS can leverage advanced AI capabilities to improve patient outcomes, advance data processing, and enhance decision-making.

In the era of Large Language Models (LLMs), the application of advanced AI technologies to data captured by wearables devices, combined with the fusion of contextual data, presents a revolutionary approach to enhancing real-time public safety, individual security, and emergency response. In this paper, we introduce WatchOverGPT, a novel framework that leverages this integration to promptly identify and respond to potential life-threatening criminal activities and safety concerns. WatchOverGPT combines the capabilities of wearable cameras, smartphones' location data, and LLM-based advanced con-versational AI communication through Generative Pre-trained Transformer (GPT). The core of this framework involves a wearable camera connected to the user's smartphone, which continuously captures and analyzes the environment for signs of distress or criminal behaviors, including human actions and the presence of weapons, coupled with location and other information from the smartphone by which GPT-based application provides an autonomous decision-making process. This paper explores the framework's design, implementation, and potential impact of LLM applications on public safety. The proposed framework aims to bridge the gap between safety threats and emergency response teams in the fight against crime through real-time data processing and AI -driven autonomous communication, enhancing the security of individuals in various settings,

The recent advancement of pretrained models shows great potential as well as challenges for privacy-preserving distributed machine learning technique called Federated Learning (FL). With the growing demands of foundation models, it is now an urgent need to explore the potential of such foundation models in a distributed setting. In this paper, In this paper, we delve into the complexities of leveraging foundation models, like CLIP into FL frameworks to preserve data privacy, and efficiently training distributed network clients across heterogeneous data landscapes. We specifically aim to address the issues related to non-IID data distributions, skewed class representation of FL clients' local dataset, communication overhead and high resource consumption due to large, complex model training in an FL setting. To address these, we propose TriplePlay, a framework that tailors CLIP foundation model as an adapter to strengthen FL model's performance and adaptability across heterogeneous data distributions among the clients. Besides, we address the long-tail distribution problem in an FL environment to maintain fairness and optimize the computational resource demands of the FL clients through quantization and low-rank adaptation techniques. A comprehensive simulations results with two distinct datasets and different FL settings demonstrate that TriplePlay efficiently reduces GPU usage and accelerates the convergence time that ultimately reduces the communication cost.

Fall incidents are a major concern for vulnerable populations, significantly contributing to annual mortality rates and highlighting the need for effective remote healthcare monitoring and response systems. Despite the introduction of numerous AI-based automatic fall detection methodologies, existing systems lack continuous communication support and often fail to adequately serve vulnerable populations, such as disabled individuals who may be unable to communicate. This study proposes a novel fall detection system for smart homes using small language models (SLMs) to address these limitations. By incorporating real-time conversation support, the system enhances its usefulness for vulnerable populations and improves communication with emergency care teams. We present the system design, integrate lightweight machine learning models, and evaluate its performance using the TinyLlama and Phi-3 models. Our evaluation focuses on accuracy, response time, and relevance of responses, providing insights into the effective implementation of language models in high-reliability systems for smart home healthcare monitoring.

Wearable AI systems, particularly in Human Activity Recognition (HAR), are becoming integral to applications in healthcare, security, and personal fitness due to the widespread adoption of smart devices and wearable technologies. However, the increasing reliance on machine learning models in HAR introduces significant risks, especially from poisoning attacks that compromise system reliability and data integrity. This paper explores the potential of Large Language Models (LLMs) to detect and sanitize poisoning attacks in wearable AI systems. Building on ongoing research into integrating LLMs within cyber-physical systems, we focus on sensor-based interactions with the physical world. Our case study seeks to answer the following question: How effective are LLMs in detecting and sanitizing poisoning attacks on human activity sensor data? Through zero-shot learning, we evaluate the performance of models such as ChatGPT 3.5, ChatGPT 4, and Gemini, providing insights into the viability of LLMs for real-time defense and data integrity in wearable AI systems.

Differentially-Private Federated Learning (DPFL) is an emerging privacy-preserving distributed machine learning paradigm that allows for the automatic recognition of human activities using wearable sensors without compromising users’ sensitive data. However, this decentralized approach makes the system vulnerable to poisoning attacks, where malicious agents can inject contaminated data during local model training. This paper presents the results of our research on designing, developing, and evaluating a holistic model for data poisoning attacks in DPFL-based human activity recognition (HAR) systems. Specifically, we focus on label-flipping poisoning attacks, where the label of a sensor reading is maliciously changed during data collection. To investigate the impact of such attacks, we develop a simulator that explores key design issues, such as the correlation between the level of differential privacy, the level of poisoning, the number of communication rounds, and the number of agents in the system. Our findings shed light on the effectiveness of label contamination attacks in DPFL-based HAR systems and can inform the development of more robust and secure models.

The adoption of blockchain in the Internet of Things (IoT) has been increasing due to the various benefits that blockchain brings, such as security and privacy. Current blockchain models for mobile IoT assume there are fixed, powerful edge devices capable of providing global communication to all the nodes in the network. However, due to the mobile nature of IoT or network partitioning problems (NPP), nodes can move out of a cell area and split into smaller independent peer-to-peer subnetworks. Existing blockchain structures either do not support the network partitioning problem or have limitations. This paper introduces a multidimensional, graph-based blockchain structure, that utilizes k-dimensional spatiotemporal space, to address the challenges of applying blockchain in mobile networks with limited resources. Experimental results show that a multidimensional blockchain structure can improve scalability and efficiency as the blockchain grows in size, similar to logarithmic growth, and reduce the longest chain length by more than 99.99% compared to the traditional chain-based blockchain structure.

Recently, blockchain has received much attention from the mobility-centric Internet of Things (IoT). It is deemed the key to ensuring the built-in integrity of information and security of immutability by design in the peer-to-peer network (P2P) of mobile devices. In a permissioned blockchain, the authority of the system has control over the identities of its users. Such information can allow an ill-intentioned authority to map identities with their spatiotemporal data, which undermines the location privacy of a mobile user. In this paper, we study the location privacy preservation problem in the context of permissioned blockchain-based IoT systems under three conditions. First, the authority of the blockchain holds the public and private key distribution task in the system. Second, there exists a spatiotemporal correlation between consecutive location-based transactions. Third, users communicate with each other through short-range communication technologies such that it constitutes a proof of location (PoL) on their actual locations. We show that, in a permissioned blockchain with an authority and a presence of a PoL, existing approaches cannot be applied using a plug-and-play approach to protect location privacy. In this context, we propose BlockPriv, an obfuscation technique that quantifies, both theoretically and experimentally, the relationship between privacy and utility in order to dynamically protect the privacy of sensitive locations in the permissioned blockchain.

Mobile crowdsensing, an emerging sensing paradigm, promotes scalability and reduction in the deployment of specialized sensing devices for large-scale data collection in a decentralized fashion. However, its open structure allows malicious entities to interrupt a system by reporting fabricated or erroneous data, making trust evaluation a highly important issue in mobile crowdsensing applications. The goal of this research is to show that an introduction of a reputation system in the process of correlated sensor-based data fusion will enhance the overall quality of the sensed data. To do so, we design a reputation-aware data fusion mechanism to ensure data integrity. We use Gompertz function in our reputation method to rate the trustworthiness of the data reported by a crowdsensing participant. The proposed mechanism, on one hand, is capable of defending a data corruption attack and identifying malicious or honest participants based on their reported data in real time. On the other hand, this mechanism yields more accurate data prediction in terms of lower data prediction error. We conducted experiments using two different real-world datasets. We compare our correlated data and reputation-aware data prediction (CDR) method with other popular methods, and the results show that our effective method incurs lower data prediction error.

In Location-Based Services (LBS), users are required to disclose their precise location information to query a service provider. An untrusted service provider can abuse those queries to infer sensitive information on a user through spatio-temporal and historical data analyses. Depicting the drawbacks of existing privacy-preserving approaches in LBS, we propose a user-centric obfuscation approach, called KLAP, based on the three fundamental obfuscation requirements: k number of locations, l-diversity, and privacy area preservation. Considering user's sensitivity to different locations and utilizing Real-Time Traffic Information (RTTI), KLAP generates a convex Concealing Region (CR) to hide user's location such that the locations, forming the CR, resemble similar sensitivity and are resilient against a wide range of inferences in spatio-temporal domain. For the first time, a novel CR pruning technique is proposed to significantly improve the delay between successive CR submissions. We carry out an experiment with a real dataset to show its effectiveness for sporadic, frequent, and continuous service use cases.