The fast growth of the IoT systems has made networks highly vulnerable to identity-based cyberattacks, and thus intrusion detection system (IDS) is necessary to be accurate and scalable. The current paper is a systematic analysis of ML and ensemble model-based approaches to identify IoT identity attacks on the CICIDS2017 data. An end-to-end pipeline is developed which is reproducible, including data cleaning, feature standardization, label encoding, and supervised learning. Logistic Regression, Decision tree, random Forest, and XGBoost classifiers are assessed using the following metrics: accuracy, precision, recall, F1-score, and ROC-AUC. As shown by testing results, XGBoost outperforms both the baseline DL model and the classic ML method, boasting an accuracy of 99.89 and a flawless ROC-AUC of $\mathbf{1. 0 0}$. Besides the general performance, a comprehensive analysis of attacks and errors is processed, which shows that major attacks, such as DDoS, DoS variants, PortScan, and Heartbleed, are detected almost perfectly, and minor and behaviorally overlapping attacks, including Web AttackXSS and Infiltration, continue to pose significant difficulties. The presented results are useful in practice as demonstrate significance of features, the effect of imbalanced classes, and trends of misclassification and help to create effective and scalable IDS systems in real IoT scenarios.
Low Earth orbit (LEO) satellite systems equipped with large-scale multiple-input multiple-output (MIMO) antenna arrays have emerged as a key enabler for next-generation Earth observation (EO) services. By exploiting flexible beamforming, massive-MIMO LEO satellites can support simultaneous data delivery to multiple ground users, while accommodating the stringent quality-of-service (QoS) requirements of EO applications. However, the rapidly growing volume of EO data, together with limited onboard power and highly dynamic satellite channels, poses significant challenges for efficient data transmission. To address these challenges, this paper investigates a SemCom framework based on deep joint source-channel coding (DJSCC), which directly maps sensed EO data into channel symbols and prioritizes the transmission of task-relevant information. Within this framework, we study a resource allocation problem that jointly optimizes transmission power and semantic compression to maximize the minimum QoS across users, defined in terms of image reconstruction quality. By leveraging a curve-fitting model that characterizes the relationship among compression ratio, signal-to-interference-plus-noise ratio (SINR), and semantic reconstruction quality, the proposed approach enables a tractable solution for massive-MIMO LEO satellite systems. Numerical results demonstrate that the proposed framework effectively improves EO transmission performance while efficiently utilizing limited satellite resources.
We introduce Sparse Selective Hyper-Connections (SHC), a novel residual connection paradigm that addresses critical limitations of Manifold-Constrained Hyper-Connections (MHC) while maintaining their stability guarantees. Our key insight is that the Birkhoff-von Neumann theorem permits exact representation of doubly stochastic matrices as sparse convex combinations of permutation matrices, eliminating the need for iterative Sinkhorn normalization. We synthesize concepts from six mathematical domains―convex geometry, Lie groups, random projections, state space models, differentiable sorting, and tensor decomposition―to construct an architecture that achieves: (1) closed-form orthogonal routing via the Cayley transform, (2) $\mathcal{O}(1)$ memory scaling through rank- $r$ factorization inspired by Johnson-Lindenstrauss, and (3) inference-time efficiency via SSM distillation. Theoretical analysis proves SHC maintains bounded spectral norm $\rho\left(\mathbf{H}^{\text {res }}\right)=1$ exactly, compared to $\rho \leq 1.6$ for MHC. Our framework reduces KV cache overhead from $4 \times$ to $\sim 1.2 \times$ while preserving expressivity, enabling practical deployment of multi-stream residual architectures at scale.
Traditional security questions (e.g., mother's maiden name or first school) are easy to implement and use, but vulnerable to adversarial guessing or research, due to the increasing availability of personal information online, which reduces their effectiveness in cybersecurity authentication. Choice-based security questions offer a promising alternative by leveraging stable personal preferences that are harder for adversaries to predict. Prior studies have not comprehensively examined the temporal stability and consistency of users' responses to a wide range of choice-based questions and their potential to improve the authentication. In this work, we study and analyze responses from 276 participants to nine choice-based questions (e.g., favorite color, number, season, day, animal, sport, month, food, and artist), collected three times with one-month intervals. Results show that questions on color, number, season, day, and animal exhibit high stability over a 2 -month period, while maintaining high uncertainty against adversarial guessing, making them reliable candidates for robust authentication. In contrast, sport, month, food, and artist questions demonstrate lower stability. These findings highlight the importance of balancing stability (in terms of user memorability and recall consistency) with uncertainty (in terms of resistance in adversarial prediction) when designing choice-based questions, enabling more secure, user-friendly authentication mechanisms in cybersecurity systems.
Modern methods of environmental monitoring are deficient in the lack of ability to take measurements of energy flows since traditional readings involve capturing parameters such as temperature, pressure, and humidity without considering their physical causes. The present research describes Differential Temporal Derivative Soft-Sensing (DTDSS), a physics-based approach which enables any ordinary low cost sensor array to infer estimates of the energy exchange in the environment by modeling its radiative heat fluxes. In particular, the proposed approach combines a novel paired sensor configuration along with a unique algorithmic solution called Inertial Noise Reduction or INR, that mathematically models the flow of energy in the environment by computing Global Horizontal Irradiance, or GHI, and convective heat flux. Experimental field testing has been conducted with the use of calibrated reference pyranometers supplied by the Department of Meteorology of Sri Lanka, yielding a correspondence between 8 bit embedded processor results and the reference of R2 approx. eqv. to 0.9 and RMSE approx. eqv. to 45 Watts per square meter in under 2KB RAM of a microcontroller unit.
Reliable grid operation depends on accurate and timely telemetry, making modern power systems vulnerable to communication layer cyberattacks. This paper evaluates how Denial of Service (DoS), Denial of Data (DoD), and False Data Injection (FDI) attacks disrupt the IEEE 14 bus system using a MATLAB only, time stepped simulation framework built on MATPOWER. The framework emulates a 24 hour operating cycle with sinusoidal load variation, introduces attack specific manipulation of load and voltage data, and performs full AC power flow solves with reactive limit enforcement (PV PQ switching). At each timestep, the system logs true and measured voltages, generator P/Q output, system losses, and voltage limit violations to capture transient cyber physical effects. Results show that DoD causes the largest physical distortions and reactive power stress, DoS masks natural variability and degrades situational awareness, and FDI creates significant discrepancies between true and perceived voltages. The study provides a compact, reproducible benchmark for analyzing cyber induced instability and informing future defense strategies.
Wi-Fi deauthentication attacks remain a practical denial-of-service (DoS) threat by exploiting unprotected management frames to disrupt client connectivity. In this work, we introduce a software-defined testbed to measure Wi-Fi resilience to deauthentication attacks. We experimentally evaluate five wireless security configurations: open networks, WPA1, WPA2 without Protected Management Frames (PMF), WPA2 with PMF, and WPA3. Using controlled experiments, we measure client disconnection rates, packet injection volume, and time-to-disruption under each configuration. Packet-level behavior is analyzed using standard wireless auditing tools. Open networks, WPA1, and WPA2 without PMF proved entirely vulnerable to deauthentication, while no successful attacks were observed for WPA2 with PMF or WPA3 under tested conditions. These findings confirm the effectiveness of management-frame protection and highlight the continued risk posed by legacy or misconfigured wireless deployments.
As large language models (LLMs) advance toward expert-level performance in engineering domains, reliable reasoning under user-specified constraints becomes critical. In circuit analysis, for example, a numerically correct solution is insufficient if it violates established methodological conventions such as mesh directionality or polarity assignments, errors that can propagate in safety-critical systems. Yet it remains unclear whether frontier models truly apply first-principles reasoning or rely on entrenched training priors that conflict with explicit instructions. We introduce CircuChain, a diagnostic benchmark designed to disentangle instruction compliance from physical reasoning competence in electrical circuit analysis. CircuChain consists of counterbalanced Control/Trap problem pairs across five canonical circuit topologies, augmented with systematic variations in sign conventions, current orientations, and polarity definitions. A multi-stage verification pipeline, combining symbolic solvers, SPICE simulation, and an LLM-based error taxonomy, enables fine-grained attribution of failures to convention errors, physics errors, arithmetic mistakes, or hallucinations. Across 100 tasks per model, we observe a consistent Compliance-Competence Divergence. The strongest model evaluated exhibits near-perfect physical reasoning but a high rate of convention violations when Trap conditions deliberately invert natural sign patterns. Conversely, weaker models display lower physical fidelity yet superior adherence to explicit instructions. These results suggest that increased model capability does not guarantee improved constraint alignment and highlight the need for new evaluation frameworks that stress instruction-following under mathematically rigid domains. CircuChain provides one such framework and offers actionable insights for both engineering education and AI alignment research.
Though Explainable AI (XAI) has made significant advancements, its inclusion in edge and IoT systems is typically ad-hoc and inefficient. Most current methods are "coupled" in such a way that they generate explanations simultaneously with model inferences. As a result, these approaches incur redundant computation, high latency and poor scalability when deployed across heterogeneous sets of edge devices. In this work we propose Explainability-as-a-Service (XaaS), a distributed architecture for treating explainability as a first-class system service (as opposed to a model-specific feature). The key innovation in our proposed XaaS architecture is that it decouples inference from explanation generation allowing edge devices to request, cache and verify explanations subject to resource and latency constraints. To achieve this, we introduce three main innovations: (1) A distributed explanation cache with a semantic similarity based explanation retrieval method which significantly reduces redundant computation; (2) A lightweight verification protocol that ensures the fidelity of both cached and newly generated explanations; and (3) An adaptive explanation engine that chooses explanation methods based upon device capability and user requirement. We evaluated the performance of XaaS on three real-world edge-AI use cases: (i) manufacturing quality control; (ii) autonomous vehicle perception; and (iii) healthcare diagnostics. Experimental results show that XaaS reduces latency by 38% while maintaining high explanation quality across three real-world deployments. Overall, this work enables the deployment of transparent and accountable AI across large scale, heterogeneous IoT systems, and bridges the gap between XAI research and edge-practicality.
Large language models (LLMs) are increasingly applied in mental health support systems, where reliable recognition of high-risk states such as suicidal ideation and self-harm is safety-critical. However, existing evaluations primarily rely on aggregate performance metrics, which often obscure risk-specific failure modes and provide limited insight into model behavior in realistic, multi-turn interactions. We present MHDash, an open-source platform designed to support the development, evaluation, and auditing of AI systems for mental health applications. MHDash integrates data collection, structured annotation, multi-turn dialogue generation, and baseline evaluation into a unified pipeline. The platform supports annotations across multiple dimensions, including Concern Type, Risk Level, and Dialogue Intent, enabling fine-grained and risk-aware analysis. Our results reveal several key findings: (i) simple baselines and advanced LLM APIs exhibit comparable overall accuracy yet diverge significantly on high-risk cases; (ii) some LLMs maintain consistent ordinal severity ranking while failing absolute risk classification, whereas others achieve reasonable aggregate scores but suffer from high false negative rates on severe categories; and (iii) performance gaps are amplified in multi-turn dialogues, where risk signals emerge gradually. These observations demonstrate that conventional benchmarks are insufficient for safety-critical mental health settings. By releasing MHDash as an open platform, we aim to promote reproducible research, transparent evaluation, and safety-aligned development of AI systems for mental health support.
Automated negotiations in insurance and business-to-business (B2B) commerce encounter substantial challenges. Current systems force a trade-off between convenience and privacy by routing sensitive financial data through centralized servers, increasing security risks, and diminishing user trust. This study introduces a device-native autonomous Artificial Intelligence (AI) agent system for privacy-preserving negotiations. The proposed system operates exclusively on user hardware, enabling real-time bargaining while maintaining sensitive constraints locally. It integrates zero-knowledge proofs to ensure privacy and employs distilled world models to support advanced on-device reasoning. The architecture incorporates six technical components within an agentic AI workflow. Agents autonomously plan negotiation strategies, conduct secure multi-party bargaining, and generate cryptographic audit trails without exposing user data to external servers. The system is evaluated in insurance and B2B procurement scenarios across diverse device configurations. Results show an average success rate of 87
Large Language Models (LLMs) such as ChatGPT-4, Claude 3, and LLaMA 4 are increasingly embedded in software/application development, supporting tasks from code generation to debugging. Yet, their real-world effectiveness in detecting diverse software bugs, particularly complex, security-relevant vulnerabilities, remains underexplored. This study presents a systematic, empirical evaluation of these three leading LLMs using a benchmark of foundational programming errors, classic security flaws, and advanced, production-grade bugs in C++ and Python. The dataset integrates real code from SEED Labs, OpenSSL (via the Suresoft GLaDOS database), and PyBugHive, validated through local compilation and testing pipelines. A novel multi-stage, context-aware prompting protocol simulates realistic debugging scenarios, while a graded rubric measures detection accuracy, reasoning depth, and remediation quality. Our results show that all models excel at identifying syntactic and semantic issues in well-scoped code, making them promising for educational use and as first-pass reviewers in automated code auditing. Performance diminishes in scenarios involving complex security vulnerabilities and large-scale production code, with ChatGPT-4 and Claude 3 generally providing more nuanced contextual analyses than LLaMA 4. This highlights both the promise and the present constraints of LLMs in serving as reliable code analysis tools.
In this paper we introduce a novel dataset of high altitude aerial images with manual annotations for impervious surfaces from a predominantly urban area. The dataset also contains a small amount of hand annotations for man-made pervious surfaces, such as unpaved road and driveways. Each aerial image also has unlabeled regions that are neither impervious nor man-made pervious. A pervious surface is defined as one that allows water to pass through whereas an impervious one does not allow water to percolate. After introducing the dataset, we study the problem of automatically categorizing the regions in an aerial image into their respective categories. To this end we explore three directions: 1) predicting impervious, pervious and unlabeled (background), 2) predicting impervious and the combined class of unlabeled and man-made pervious and 3) predicting impervious and background excluding man-made pervious regions. For each of these problems, we used a custom built UNet architecture with dropout layers. We report the results from the same in the form of classification masks. The results were verified by manually aligning the predicted mask with the ground truth mask for impervious, man-made pervious, and background regions. Our experiments and results establish the efficacy of using deep learning techniques for the problem of automatically identifying impervious surfaces from high-resolution aerial imagery, especially in complex urban areas.
Limb loss and stroke-related motor impairments significantly impact individuals' mobility and independence, demanding advanced prosthetic and rehabilitation technologies. Surface electromyography has been widely used for prosthetic control, enabling gesture recognition through machine learning and deep learning models. However, sEMG-based systems often struggle with signal reliability and limited resolution for complex hand movements. Ultrasound has been explored as a complementary sensing modality, providing deeper muscle activity insights. This study proposes a novel deep learning approach integrating sEMG and US signals to enhance continuous 2-DoF hand motion estimation. We employ a sensor fusion technique utilizing an 8-channel sEMG armband and a handheld US probe, alongside an inertial measurement unit for ground truth data collection. Our method employs a hybrid convolutional neural network and long-short-term memory model to capture both spatial and temporal features of muscle activity, improving accuracy in predicting dynamic hand gestures. Experimental results suggest that the fusion of sEMG and US enhances continuous estimation accuracy particularly in fast and complex hand movements, compared to single-modality approach. The proposed method shows a potential of integrating hybrid sensing and deep learning for more intuitive and responsive prosthetic control systems.
NASA's Planetary Data System (PDS) is a crucial repository for planetary mission data, yet its complex interface challenges users, especially non-experts. This research develops a framework that leverages OpenAI's GPT-4o-mini-API and a Retrieval-Augmented Generation (RAG) with the aim of simplifying user interaction with PDS. Using semantic embeddings, BM25 retrieval, and GPT-4o-mini, the system translates natural language queries into structured outputs, such as PDS compatible URLs. Our pilot implementation testing with simple queries achieved 95.24% accuracy in generating correct URLs, outperforming traditional TF-IDF with cosine similarity approach, which gained 90.91% accuracy. This work highlights the transformative potential of large language models to improve usability for scientific data repositories.
We investigate a method to utilize source memory information for a Soft-Output Viterbi algorithm (SOVA) decoded Serially Concatenated Continuous Phase Modulation (SCCPM) system. The system concatenates a recursive systematic convolutional (RSC) code with a binary full-response continuous phase modulation (CPM) signal. We apply the SOVA as the Soft-Input Soft-Output (SISO) decoder for both the CPM and RSC components. We modify the RSC SOVA decoder to combine the a priori information provided by the CPM SOVA decoder with the a priori information from a Markov source. Simulation results show that incorporating source statistics information improves the error performance of the SCCPM system and accelerates decoding convergence during iterative decoding.
In today's digital age, a key challenge is the need for secure and reliable identity verification. Traditional methods, often relying on centralized authorities, are vulnerable to data breaches, privacy concerns, and inefficiency. For example, the 2017 Equifax breach exposed the personal data of over 147 million individuals, highlighting the risks of centralized data storage. Similarly, traditional identity verification methods in banking often involve lengthy processes prone to human error and fraud. These methods pose risks such as identity theft and lack of user control over personal data; for instance, studies show that identity theft affected over 42 million individuals in the U.S. in 2021 alone, highlighting the critical need for improved security measures. Therefore, we need to design a robust security and highly scalable framework to effectively address user privacy concerns. To address these challenges, we propose a blockchain-based digital identity verification system that leverages the immutability and transparency of blockchain technology. Our research provides a comprehensive discussion on the system's architecture, implementation, and potential challenges, paving the way for secure and decentralized identity verification frameworks. Our framework enhances data security, ensures user privacy, and fosters trust between stakeholders. The experiment's results demonstrate significant improvements in efficiency and enhanced security compared to centralized solutions.
Wheelchair users with limited or nonexistent lower-limb mobility often struggle with significant muscle atrophy, circulatory complications, and other health issues. Although exoskeletons for gait training have shown promise, many are optimized for upright ambulation and overlook individuals remaining fully seated due to clinical or practical constraints. This article presents a seat-focused exoskeleton synergy integrated with a wheelchair to facilitate multi-DOF lower-limb mobilization without requiring transfers or standing maneuvers. The system addresses a critical gap by providing repeated, lower-intensity knee and ankle exercises throughout the day-either passively, actively with assistance, or in a resistive training mode-thus promoting muscle engagement and circulation from a seated position. The exoskeleton leverages the wheelchair's power source and employs modular, adjustable rails to accommodate diverse leg lengths. Emphasis is placed on biomechanical alignment, safe torque control, and user comfort, supported by a model-based controller accounting for inertia, Coriolis effects, and gravity in real time. Bench tests and preliminary trials confirm stable torque output, accurate tracking of desired motion profiles, and minimal burden when attaching or detaching the device. The article introduces modeling for a two-DOF prototype (knee flexion/extension and ankle dorsiflexion/plantarflexion), expandable to include additional joints or degrees of freedom, enabling broader rehabilitation targets such as hip abduction/adduction. By embedding rehabilitative capabilities directly into the wheelchair environment, the proposed approach seeks to close the gap between infrequent clinic appointments and continuous therapy aimed at mitigating muscle ailments. A seat-focused exoskeleton designed for every-day use can facilitate accessible, on-demand rehabilitation and substantially improve long-term health outcomes for mobility impaired users.
Real-time object detection is pivotal in various computer vision and robotics applications, such as security surveillance, autonomous navigation, and interactive systems. This study evaluates several object detection algorithms under low illumination conditions, comparing traditional methods including SIFT, SURF, ORB, BRISK, KAZE, and AKAZE with the deep learning model YOLOv5. To simulate these conditions, we utilized the COCO dataset and implemented a sequence of preprocessing procedures, such as adjusting the gamma correction throughout the dataset, introducing Gaussian noise, and equalizing the histogram to improve image contrast. Here, YOLOv5 exhibited superior performance, achieving a detection accuracy of 92.47% and a processing time of 45 milliseconds per image. Traditional algorithms like SURF and KAZE, although slower, demonstrated precision rates of 78.54% and 76.86%, respectively. YOLOv5 also identified the highest number of key points and matches. Our findings indicate that while YOLOv5 excels in real-time detection under low-light conditions, traditional methods are effective for specific tasks in resource-constrained environments. This study benchmarks these algorithms and proposes a hybrid approach to enhance real-time object detection systems under low illumination conditions.
The fidelity and robustness of the models is critical for the efficacy of modern distribution system planning studies, serving as the foundation for strategic decision-making. However, systemic inconsistencies, incomplete datasets, and improper model data structuring introduce significant computational and operational complexities in power flow simulations, which are at the foundation of power system analyses and planning studies (such as hosting capacity evaluations). This paper presents a comprehensive analysis of prevalent inaccuracies and data anomalies that pervade distribution system models, categorizing them based on their impact on network topology, electrical parameter accuracy, and computational feasibility. The study investigates connectivity discrepancies, absent or misclassified asset characteristics, erroneous transformer-load associations, and data schema mismatches, elucidating their potential to distort system behavior and compromise reliability assessments. Furthermore, the paper highlights the imperative for adaptive data governance frameworks, advanced validation techniques, and automated rectification mechanisms to enhance distribution system model integrity.
