Connected and Autonomous Vehicles (CAVs) enhance the safety and efficiency of transportation by communicating sensor data through Basic Safety Messages (BSMs). This coordinated navigation strategy is vulnerable to both unavoidable sensor faults and intentional malicious attacks. Confusing sensor faults and malicious attacks can lead to inappropriate responses that compromise the safety of passengers and infrastructure. This work addresses this issue by first systematically introducing both simulated faults (drift, hard-over) and malicious attacks (colluding and replay) into the publicly-available Tampa CV Pilot BSM dataset. We then evaluate the ability of various classical machine learning approaches and ensemble classifiers like Random Forest (RF) to automatically detect anomalies and distinguish between sensor faults from malicious attacks.

As cyber-physical systems (CPS) increasingly operate alongside humans in sensitive and safety-critical environments, maintaining user trust, transparency, and emotional awareness has become integral to secure and resilient system design. Emotion recognition and affect-aware interaction models provide the foundation for building interpretable and trustworthy interfaces in domains such as healthcare robotics, and smart home environments. Traditional emotion lexicons such as manual or static word emotion mappings have proven valuable but lack scalability to modern conversational and human-machine interaction contexts. This paper extends a Mixed-Level Feed-Forward Network (MLFFN) based lexicon-generation framework to construct interpretable, data-driven emotion lexicons for conversational environments. Using FastText embeddings and three benchmark datasets (EmpatheticDialogues, GoEmotions, and MELD), our approach generates a binary joy–sadness lexicon optimized for human-CPS communication. By integrating interpretable emotional cues into CPS interfaces, the proposed framework strengthens human trust, supports explainable AI (XAI) mechanisms, and enhances the safety and usability of CPS ecosystems.

Large Language Models (LLMs) are increasingly integrated into communication-centric and networked AI systems, supporting applications such as edge/cloud services, mobile assistants, and distributed collaborative platforms. However, their tendency to generate hallucinations (factually incorrect yet linguistically plausible content) poses significant risks to trust, reliability, and security in these environments. We propose SRF, a Sequential Revision Framework for paragraph-level hallucination detection and correction in pervasive and networked AI workflows. SRF decomposes text into sentences, generates fact-checking queries, retrieves supporting evidence via web-scale APIs, and applies a semantic agreement mechanism to ensure globally coherent revisions. To evaluate factual correctness, we compare SRF against two baselines—(i) direct LLM correction (without external evidence) and (ii) sentence-level RARR—on the Natural Questions and SQuAD datasets using GPT-3.5, Mistral, and Gemini models. SRF yields up to a 15% improvement in factual accuracy, with minimal disruption to writing style and paragraph coherence. By enforcing fact-consistent and trustworthy LLM outputs, SRF strengthens the integration of AI/ML into communication networks, mobile computing, and distributed cloud/edge services, reducing vulnerabilities to misinformation propagation and adversarial manipulation.

Autonomous vehicles (AVs), critical for future intelligent transportation, owe their advanced capabilities to reinforcement learning, which facilitates their intelligent decision-making. As AV adoption increases, concerns remain about their behavior in situations with moral uncertainty. The limitation of operating in limited environments and the challenge of determining credence values for ethical theories limits the moral uncertainty of AVs. This paper incorporates comprehensive exploration of new moral theory and scenario into simulation frameworks that can help overcome the state of the art limitations. We introduce justice theory inspired by the moral machine framework in Uber research platform to study the role of fairness among individuals in morally uncertain situations. Furthermore, we introduce novel reward structures in the framework similar to deontological and utilitarian theory to comprehensively evaluate with the state of art voting methodologies namely Nash voting and Variance voting. It was found that the variance voting system was effective across both sequential and nonsequential environments, while Nash voting was suitable primarily for sequential settings.

Autonomous systems, especially those in safety-critical applications like Autonomous Vehicles (AVs), require human-like reasoning capabilities to make ethical decisions. Large Language Models (LLMs) have shown potential in simulating diverse human moral responses, offering insights into how different moral frameworks, such as utilitarianism and deontological ethics, could enhance decision-making algorithms in AVs. Existing research indicates that LLMs tend to align with commonsense morality in morally unambiguous cases, but face challenges in providing detailed justifications for their choices. Studies leveraging frameworks like the Moral Machine and Moral Foundations Theory have explored how LLMs simulate human preferences. Despite this progress, a significant gap remains in understanding how gender and age impact moral preferences when decisions are influenced by LLMs in autonomous systems. This paper addresses this gap by investigating how LLM-based systems can reflect and adapt to moral preferences across gender and age groups, while ensuring that these systems offer transparent explanations that align with human moral intuitions in high-stakes AV decision-making scenarios. It was found that LLM models demonstrated diverse tendencies: some leaned towards favoring younger individuals over older ones, while others displayed a subtle preference for males in decision-making situations, highlighting differences in how the models prioritized age and gender.

Previous research on the ethical decision-making capabilities of large language models (LLMs) has largely relied on classification-based approaches, focusing on binary judgments of moral acceptability. While prior findings indicate strong alignment between LLMs and human judgments in morally unambiguous scenarios, the reasoning processes behind these decisions remain underexplored. This paper introduces an explainability-driven workflow to assess LLM moral reasoning by prompting models to provide both binary ethical judgments and accompanying justifications. We evaluate the semantic similarity of these justifications across varied prompting styles using latent semantic analysis. Gemini achieved the most consistent moral justification similarity across frameworks, with LLaMA3 close behind and LLaMA2 and Mistral showing greater variability, especially in deontology.

Data falsification attack in connected vehicles (CV) refer to the manipulation or alteration of data within the vehicle's communication systems. This paper discusses the critical challenges in ensuring the security of CV networks where vehicle data integrity is paramount to prevent data falsification. Various existing solutions, such as machine learning and reputation-based approaches, have limitations in terms of scalability and robustness. To address these issues, we propose a novel multi-level Dempster-Shafer with reinforcement learning (RL)-based reputation system for CV networks. We use decentralized validation that combines self and peer reports of vehicles along with centralized feedback from road side unit, merging reputation-based trust management with Deep RL. By incorporating a multi-level Dempster-Shafer model, we elevate prediction accuracy and reward values while dynamic RL optimizes the process of reputation updates.

Reinforcement learning in the field of artificial intelligence has seen tremendous advances in recent years, but there are still several limitations standing in the way of its wider practical application, including sample inefficiency, generalization, and exploration-exploitation trade-off. Deep Q-Networks (DQN) have improved the performance of RL by using deep neural networks to approximate the Q-function and by using experience replay to store and reuse past experiences. Episodic memory in RL is a technique that allows an agent to store and reuse past experiences in order to improve its decision-making. However, current episodic memory-based RL techniques have some issues, such as generalization and slow learning, which can be improved by using methods such as experience replay compression and reducing the information of episodic memory into a parametric model. In this work, we propose cognitive-inspired episodic memory in DQN networks (CEMDQN) that reduces the priority weighting of old experiences over time, and the agent accesses recent experiences more frequently. The proposed model was evaluated on three different environments: StarGunner, BattleZone, and TimePilot. It was shown that when compared to standard episodic memory DQN, CEMDQN was more effective in test score performance for StarGunner (45.1 %), BattleZone (81 %), and TimePilot (53.54%) environments, respectively.

While centralized security methods such as Public Key Infrastructure (PKI) have fortified connected vehicles (CVs) security, decentralized approaches show promise in distributing security responsibilities and reducing vulnerability to cyber threats. Implementing a decentralized architecture can decrease susceptibility to large-scale cyber-attacks and enhance overall system robustness. Consensus-based trust models present avenue for scalable, decentralized security protocols in CV networks. While numerous consensus-based trust algorithms have been proposed, their validation often relied on non-representative datasets, prompting the development of more accurate simulation methodologies. This research enhances the simulation workflow by diversifying noise models, using a benchmark consensus-based trust algorithm, and addressing persistent attack model limitations.

Connected vehicles (CVs) offer enhanced safety features and improved traffic management capabilities but face a critical concern---vulnerability to malicious attacks due to interconnectivity. To address this, trust algorithms have been developed to assess vehicle trustworthiness. However, in majority-malicious conditions, these algorithms may fail. This paper explores an algorithm based on the DeGroot opinion dynamics model to achieve consensus in such scenarios, aiming to distinguish between benevolent and malicious vehicles. The model successfully identified benevolent and malicious vehicles in up to 98% corruption with an average F1 score of 0.96 on three different datasets namely motion model, open source traffic simulator, and real-world dataset.

This paper introduces a new approach to incentivise the vehicles in connected vehicle (CV) networks based on the reputation measures along with a combination of graph neural network and reinforcement learning (GNN-RL). The proposed method enables vehicles to create reputation estimates of their nearby vehicles by analyzing broadcasted kinematic data and onboard sensor estimates, as well as the network connectivity topology. This data is then utilized to create a graphical representation of reputation distribution. A centralized RL agent is used for providing reward signals to each vehicle based on a Laplacian matrix, which encourages the vehicles to make more accurate reputation estimates. The proposed algorithm is based on a GNN-RL algorithm previously used for coordinated navigation, which has been adapted to the cybersecurity domain in this paper. The simulation results show that the model was effective in giving dynamic rewards to vehicles based on their reputation scores. Further, Laplace matrices helped in analyzing the connectivity and behavior of CVs in the network.

Reinforcement learning (RL) algorithms employ a fixed-ratio schedule which can lead to overfitting, where the agent learns to optimize for the specific rewards it receives, rather than learning the underlying task. Further, the agent can simply repeat the same actions that have worked in the past and do not explore different actions and strategies to see what works best. This leads to generalization issue, where the agent struggles to apply what it has learned to new, unseen situations. This can be particularly problematic in complex environments where the agent needs to learn to generalize from limited data. Introducing variable reward schedules in RL inspired from behavioral psychology can be more effective than traditional reward schemes because they can mimic real-world environments where rewards are not always consistent or predictable. This can also encourage an RL agent to explore more and become more adaptable to changes in the environment. The simulation results showed that variable reward scheme has faster learning rate as compared to fixed rewards.

Connected Vehicles (CVs), an integral part of the future of intelligent transportation systems, use communication and sensing technologies to communicate among vehicles and infrastructure. However, as vehicles become interconnected, the vulnerability of their components to anomalies and deliberate malicious activity increases. In both cases, it is vital to detect and exclude anomalous data from the decision-making process. While deep learning techniques are gaining popularity for anomaly detection due to their adaptability, they are computationally expensive and require long training times. To overcome this challenge, this paper uses a directed acyclic graph (DAG) based distributed ledger technique and combines it with social psychology principles of ability, integrity, and benevolence to calculate the reputation of vehicles. We introduce the probability of malevolence, a measure of quality, which is a function of the error measurements (between ground truth and reported values) and reputation metrics. We introduce various anomalies such as bias, noise, short, multi-short, drift, multi-drift, stuck-at, and parasite chain attack in the simulated data from the Intelligent Driver Module framework on road topology such as uphill, ring, on-ramp, off-ramp, and road-works to validate the efficacy of the proposed framework in identifying the anomalies. Simulation results show that the malevolence factor serves as an efficient metric for automatically determining the types of anomalies in the CV network.

As recent advances in sensing, computing, and communications expedite proliferation of autonomous vehicles (AV), their sharing the road with human driven vehicles presents a challenge that demands urgent investigation. AVs can excel at deterministic programmed behavior, still human drivers have the edge because of the faculty of cognition, which evolved over millennia. This paper presents Cognition Aided Reinforcement Learning (CARL) algorithm that harnesses inputs from five principles of cognition — memory, attention, language, perception, and intelligence. Sensors build perception, data facilitate memory, and safety messages enable language support. Intelligence fuses information with attention focused on specific actions for reward maximization. Simulation results show CARL to be 10 times faster as compared to the state of the art model-free reinforcement learning algorithms. Additionally, by using the principle of metacognition (art of learning how to learn), CARL achieves optimal rewards in a heterogeneous environment composed of vehicles with varying degrees of autonomy.

Connected vehicles are set to define the future of transportation; however, this upcoming technology continues to be plagued with serious security risks. If these risks are not addressed in a timely fashion, then they could threaten the adoption and success of this promising technology. This article deals with a specific class of attacks in connected vehicles, namely tampering attacks caused due to compromise of on-board sensors. Current centralized solutions that employ trusted infrastructure to protect against adversarial manipulation of information cannot validate the correctness of the shared data and do not scale well. To overcome these issues, decentralized protection mechanisms by means of blockchain technology have emerged as a promising research direction. However, current permission-less, linear blockchain-based solutions have low transaction performance and high computational cost, thereby making it difficult to adopt them for security in connected vehicles. In this article, we present TangleCV, a directed acyclic graph–based distributed ledger technique for connected vehicles to address data tampering threats in connected vehicular networks. We describe new validation steps, tip selection strategies, and cumulative weight definition for TangleCV that not only meets the timing constraints of the connected vehicular networks but also secures the network against threats due to tampering attacks. We describe how the reputation of the network is established in TangleCV using trust factors calculated on the basis of ability, integrity, and benevolence of the nodes in the network. We present numerical results that demonstrate that the average value of the time to first approval decreases by more than 70% as the network evolves from a low load to a high load in the case of the nearest neighbor strategy. We observe that more than 60% of the nodes are approved in a low-load network and this number increases to 80% in a high-load network for the nearest neighbor strategy. The standard deviation of error measurements for nodes experiencing tampering attack is around 60% higher as compared to nodes that do not experience such an attack.

Wireless medical devices play an important role in providing safety and privacy to patients suffering from major health issues. These light-weight devices can be worn inside or outside the patient's body and provide more convenience and reliable doctor-patient communication. However, the design, development, and usage of these devices play a critical role in present network paradigm. They are vulnerable to network threats and attacks which break the confidentiality, integrity and availability protocols in networking scenarios. Thus, it is important to have identification and authentication of only the authorized peoplewho can operate the device. This paper proposes Dynamic Time Warping (DTW) algorithm for providing trusted authentication and identification of only authorized people using ECG signal. Here, DTW algorithm is used to measure the correlation between different ECG signal records. Experiments were carried out to evaluate the proposed algorithm with a large database consisting of users of al1 ages, including abnormal ECG data and long span of time intervals between ECG recordings for evaluating the reliability of the proposed algorithm. Comparative evaluation of the proposed sy stem show ed that, it is not only efficient, but also light weight in comparison to the existing systems.

The immune system of the human body has massive potential in defending it against multiple harmful viruses and foreign bodies. All through their developmental history, human beings have been contaminated by micro-organisms. In order to restrict the nature, size, and intensity of these microbial invasions, human beings have inherent capabilities to deal with them. The human immune system is capable of protecting the body in the form of external barriers such as skin, cells, and tissues. Furthermore, it is capable of differentiating among the self and the non-self cells with the distinct properties and features that infiltrate the human body. This paper presents a case study of the human immune system in which we develop mathematical models of innate and adaptive immune system. Extensive simulations were carried out to study the effect of the foreign particles when the recovery mechanism occurs in the body. The results obtained, substantiate the reliability of the human immune mathematical model. Finally, we advocate that having a strong security and privacy around the human body can contribute in building a strong network system. For instance, the two layer immune inspired framework viz innate layer and adaptive layer can be instigated at the data layer and the control layer of Software Defined Networking respectively.

Diabetic therapy or insulin treatment enables patients to control the blood glucose level. Today, instead of physically utilizing syringes for infusing insulin, a patient can utilize a gadget, for example, a Wireless Insulin Pump (WIP) to pass insulin into the body. A typical WIP framework comprises of an insulin pump, continuous glucose management system, blood glucose monitor, and other associated devices with all connected wireless links. This takes into consideration more granular insulin conveyance while achieving blood glucose control. WIP frameworks have progressively benefited patients, yet the multifaceted nature of the subsequent framework has posed in parallel certain security implications. This paper proposes a highly accurate yet efficient deep learning methodology to protect these vulnerable devices against fake glucose dosage. Moreover, the proposal estimates the reliability of the framework through the Bayesian network. We conduct comparative study to conclude that the proposed method outperforms the state of the art by over 15% in accuracy achieving more than 93% accuracy. Also, the proposed approach enhances the reliability of the overall system by 18% when only one wireless link is secured, and more than 90% when all wireless links are secured.

Evolution of implantable medical devices for human beings has provided a radical new way for treating chronic diseases such as diabetes, cardiac arrhythmia, cochlear, gastric diseases etc. Implantable medical devices have provided a breakthrough in network transformation by enabling and accessing the technology on demand. However, with the advancement of these devices with respect to wireless communication and ability for outside caregiver to communicate wirelessly have increased its potential to impact the security, and breach in privacy of human beings. There are several vulnerable threats in wireless medical devices such as information harvesting, tracking the patient, impersonation, relaying attacks and denial of service attack. These threats violate confidentiality, integrity, availability properties of these devices. For securing implantable medical devices diverse solutions have been proposed ranging from machine learning techniques to hardware technologies. The present survey paper focusses on the challenges, threats and solutions pertaining to the privacy and safety issues of medical devices.

Security plays a vital role in Wireless Sensor Networks (WSN) for providing reliability to the network. In WSN, where nodes, in addition to having their inbuilt capability of sensing, processing, and communicating data, also possess certain risks. These risks expose them to attacks and bring in many security challenges. Many researchers are engaged in developing innovative design paradigms to address security issues by developing trust management systems. In WSN, trust is important for the establishment of cooperation among the sensor nodes. The article presents a sociopsychological model for detecting fraudulent nodes in WSN. The three factors, viz. ability, benevolence, and integrity, are used for the computation of trust. Furthermore, the article provides a novel consensus-aware sociopsychological approach to deal even in the presence of higher number of fraudulent nodes than benevolent nodes. The proposed work has been implemented in the LabVIEW platform and extensive simulations were carried out to study its performance. Additionally, it is experimentally evaluated on a testbed of size 16 nodes to obtain results that demonstrate the accuracy and robustness of the proposed model.

Biological Immune Systems have intelligent capabilities of detecting foreign bodies which attack our body. Moreover they have inherent insightful capabilities to remember them, when they hit the body again. Primary response is the initial response instantiated by the body to the attack and secondary response is the response hence forth. Secondary response is naturally faster because of its characteristic of remembering the cure of the attack. Similar type of perspicacious nature can be adapted in removal of fraudulent nodes in wireless sensor network. The work proposes a novel algorithm for the detection and removal of the fraudulent nodes. It first detects the fraudulent nodes by machine learning module and then removes these nodes by immune-inspired module. Eventually if the same type of malicious nature is seen again, analogy of secondary response of immune system is instigated in sensor network. Proposed work has been implemented in LabVIEW platform and obtained results that demonstrate the accuracy and robustness of the proposed model.