Establishing dense, robust image correspondences under large-scale variations, perspective distortions, and occlusions remains a core challenge in computer vision. We introduce PIXELMAP, a swarm-inspired framework that refines local affine transformations in a structured Affine Correspondence Grid (AC-Grid) while ensuring global consistency through Iterative Refinement (IR). Each grid cell acts as an autonomous "agent," adapting its transformation based on a Correspondence Mapping (CM) mechanism that propagates improvements among neighbors. This localized agent-based approach naturally supports massive parallelization and achieves state-of-the-art accuracy across diverse imaging conditions-without requiring training data. Extensive experiments on synthetic distortions and real-world images demonstrate that PIXELMAP maintains high precision and recall, even under severe geometric distortions and noise. An open-source implementation and interactive website further facilitate adoption and invite contributions to this scalable, training-free solution for dense correspondence (GitHub, Interactive Website).

Semantic communications are poised to be a game changer for next-generation 6G telecommunications. However, these emerging paradigms lack a generalised model that addresses bandwidth usage, transmittable information scheduling, and semantic performance evaluation. To tackle these challenges, we theorise Differential Semantic Communications, abbreviated DSCs, inspired by human-like natural communications. A DSC model employs a closed-loop feedback architecture, where intelligent transmitter and receiver agents interact cyclically to achieve a communication goal. We introduce a top-down information representation to enable iterative transmissions. Visual salient object transmission is demonstrated as a proof-of-concept application for DSCs. In addition, a novel performance metric, termed semantic misfit-percent, is derived to evaluate the efficiency of semantic communication systems. Simulation results show the ability of salient object transmission in images with about 9 % misfit improvement compared to the state-of-the-art DALL.E1based approach on average. Implementation codes are available upon request.

The collaboration of computing powers (CPs) among unmanned aerial vehicles (UAVs)-mounted edge servers is essential to handle data-intensive tasks of user equipments (UEs). This paper presents a multi-UAV computing power network (CPN) that orchestrates the CP resources of edge servers to cooperatively process tasks through collaborative computing. Due to the frequent interactions among edge servers and the increased complexities for large-scale task executions, we propose a digital twin (DT)-driven multi-UAV CPN framework, where the aerial DT network is created as a digital replica of the multi-UAV network to assist collaborative computing. A joint optimization problem of the service association, offloading power control, task partitions, CP allocation, and UAV trajectory design is formulated to minimize the long-term average delay of UEs. We use the Lyapunov technique to tackle the long-term energy constraint, forming a tractable optimization problem, which is modeled as a Markov decision process to balance the UAV's deficit queue stability and the UE's task completion delay. A hybrid soft actorcritic based deep reinforcement learning algorithm is developed to optimize the joint design by sampling the hybrid action from the Gumbel and Gaussian distributions. Simulation results verify that the multi-UAV CPN contributes to significant performance gains when supported by DTs.

This paper addresses stability challenges in Dual Active Bridge (DAB) converters with input LC filter, increasingly adopted in emerging DC grids. The targeted system is studied in response to wide supply voltage variations in both steady-state and transient conditions. Although numerous Modulation Schemes (MSs) are developed to optimize efficiency, their impact on stability has not been fully addressed in literature. In response, a novel method for informed MS design based on small-signal stability analysis is proposed. To that end, the closed-loop input impedance of the converter is modeled with full consideration of the MS and controller parameters. The impact of an Active Damping (AD) strategy is even considered for stability margin enhancement, thus allowing for the design and experimental validation of an informed MS that strikes a balance between stability and efficiency.

In this contribution, we formalize the general case of linguistic summaries with multiple summarizers and qualifiers. We extend the definitions present in the literature to handle cases in the following form: Q R1 ⋆. .. ⋆ R k y's are P1 ⋄. .. ⋄ P l where ⋆, ⋄ ∈ {AND, OR}. Moreover, we study the consistency of fuzzy summaries, focusing mainly on the double negation property (DN). We consider different negations for linguistic quantifiers and show which can be used for preserving the DN property.

In this paper, several typical temperature sensitive electrical parameters are investigated regarding their applicability in Schottky-contact and Gate-Injection GaN HEMTs. In addition, the effect of charge carrier trapping on the temperature behavior of the threshold voltage and gate current is analyzed. It is found that the drain-source on-resistance, the drain-source capacitance, and the transconductance can all be exploited for temperature estimation. On the other hand, shifting of the threshold voltage due to gate voltage stresses obfuscates any shifting caused by temperature changes in Schottky-contact device and the ohmic-contact device to a smaller degree. The gate current of the Schottky-contact device is also stress-dependent around the threshold voltage, but stabilizes at higher gate-source voltage values. In contrast, the gate current of the ohmic-contact device is unaffected by gate stress in the transistor on-state.

To address the technical bottleneck caused by the domain mismatch problem in the sim-to-real rransfer process of humanoid robots, a targeted domain refinement (TDR) framework is proposed to solve the shortcomings of adaptive domain randomization (ADR). This research design uses quantitative simulation experiments to execute the code of TDR and ADR architecture through the Python 3.13 IDLE platform. Then the sample efficiency, adaptation speed, robustness, policy stability and domain gap of synthetic data are used to compare and analyze the two frameworks. The results show that TDR outperforms ADR on five indicators. It proves that it can make up for the shortcomings of ADR in solving domain mismatch problems, and provides reference experience for future research on improving sim-to-real rransfer.

Robust autonomous navigation in dynamic and cluttered environments is a persistent challenge for mobile robotics. Conventional methods, reliant on static maps and classical algorithms, often falter when faced with frequently reconfigured obstacle layouts. Deep Reinforcement Learning (DRL) enables adaptive, map-free navigation, yet the complexity of high-resolution LiDAR data and the presence of subtle, moving obstacles complicate training and scalability. To address these issues, we present the Multi-Head Memory Contextualising TD3 (mhmcTD3) architecture, a novel DRL framework that unites multiple specialised network heads-LiDAR fusion, LiDAR Feature, Robot States, and LiDAR Memory-within a unified actor-critic backbone. This integrated design emphasises critical LiDAR states through tailored preprocessing (inversion and exponential scaling), leverages CNNs for refined feature extraction, and employs an LSTM-based LiDAR Memory head for temporal awareness. Additionally, the incorporation of the SiLU activation function and the CoRE optimiser enhances learning stability and convergence efficiency. Extensive evaluations in both simulation (using GAZEBO with ROS2) and real-world tests on a Turtlebot3 waffle_pi platform demonstrate that mhmcTD3 delivers state-of-the-art performance under diverse LiDAR resolutions. It excels at detecting and avoiding small, dynamic obstacles while maintaining adaptability in dense, fast-changing scenarios. Ablation studies further confirm the indispensable synergy among the multiple heads. Taken together, these contributions establish mhmcTD3 as a robust and generalisable DRL-based navigation solution, bridging the gap from controlled simulations to complex real-world environments.

Detecting anomalies in a smart grid system has become critical due to the increase in cyberattacks. The fuzzy and dynamic nature of cyberattacks has caused a lot of challenges in detecting anomalies leading to abnormal predictions in a smart grid system. That has allowed cyber criminals to deploy various attacks on the network. Due to the rapidly expanding industry of cyber-physical systems, the traditional power grid has been upgraded into a communication infrastructure network. However, this upgrade has led to cyberattacks such as Ukraine power system cyberattack 2015, Iran Aramco attack 2017 and New York smart grid system attack 2018 causing a lot of damage to the systems, the economy and livelihoods. The aim of this dissertation is to explore anomaly detections in smart grid systems using machine learning models. Due to the integrated nature of the smart grid system, it has inherent vulnerabilities which attackers can exploit because of its integrated design. Such threat actors take advantage of these flaws, which has an enormous effect on embedded systems. This problem is rampant because of the potentially extensive size of these networks, which makes continuous monitoring by technicians or specialists impracticable. For this problem to be effectively resolved, automatic detection of entities that present anomalous activity is essential. Machine learning methods are important in transforming the security of smart grid systems from merely facilitating secure communication between devices to security-based intelligence systems. The dissertation proposes integrating multiple supervised machine learning-based models into a single Intrusion Detection Systems to complement each other and mitigate the shortcomings of the other models that will be suitable to detect anomalies in the smart grid communication network effectively.

In this paper, Avalanche is introduced as a robust peer-to-peer payment system that leverages the Snowball consensus protocol to facilitate secure and efficient cryptocurrency transactions. Building on the foundations laid by the Snowflake and Slush protocols, Avalanche incorporates a dynamic directed acyclic graph (DAG) to enhance transaction efficiency and security. By maintaining a single sink at the genesis vertex, the DAG intertwines transaction fates, making past decisions resistant to reversal by malicious actors. My consensus mechanism is designed to accommodate a varying presence of adversarial nodes while ensuring that virtuous transactions achieve safety and liveness guarantees. Through probabilistic decision mechanisms, Avalanche efficiently handles conflicting transactions through Snowball instances, ensuring that only valid transactions are confirmed. My findings highlight the effectiveness of the Avalanche protocol in real-world applications, presenting a significant advancement in decentralized payment systems.

The advent of multi-unmanned aerial vehicle (multi-UAV) networks in mobile edge computing (MEC) introduces dynamic computational topologies where UAVs, acting as mobile edge servers, are tasked with processing data from ground-based user equipment (UE). This paper addresses the dual challenges of optimizing both UAV deployment and task offloading within such networks to minimize communication latency and efficiently utilize UAV resources, which are limited by battery life and processing capabilities. We propose a bi-level optimization framework that simultaneously tackles the placement of UAVs and the distribution of computational tasks among them. At the higher level, UAV deployment is optimized to ensure minimal distance to the UEs, thereby reducing latency and energy consumption during data transmission. At the lower level, task offloading is optimized to balance the computational load across the UAV network, considering each UAV's capacity and battery constraints. We demonstrate through extensive simulations the significant improvements in system efficiency, latency, and resilience. This approach not only enhances the performance of UAV-assisted MEC networks but also provides scalable solutions adaptable to various operational scenarios.

As minimum area SRAM bit-cells are obtained when using cell ratio and pull-up ratio of 1, we analyze the possibility of decreasing the cell ratio from the conventional values comprised between 1.5-2.5 to 1. The impact of this option on area, power, performance and stability is analyzed showing that the most affected parameter is read stability, although this impact can be overcome using some of the read assist circuits proposed in the literature. The main benefits are layout regularity enhancement, with its consequent higher tolerance to variability, cell area reduction by 25% (with respect to a cell having a cell ratio of 2), leakage current improvement by a 35%, as well as energy dissipation reduction and a soft error rate per bit improvement of around 30%.

Accurate simulation models are essential for excavator stability control, particularly when addressing complex interactions across subsystems such as the engine and hydraulic system. The Simulink-based model used in this research incorporates a complete excavator system, including engine, hydraulic, and mechanical dynamics, requiring precise parameter estimation (PE) to handle nonlinear behaviors and unmeasurable parameters like friction coefficients and initial pressures. Traditional PE methods often struggle with these challenges, leading to inaccuracies in dynamic modeling. To address this, a novel hybrid PE framework is introduced, integrating machine learning (ML) techniques such as random forest (RF) and gradient boosting (GB) with optimization methods like gradient descent (GD) and nonlinear least squares (NLS). This framework bridges physicsbased simulations and real-time IoT-acquired data, enabling accurate estimation and correction of parameters. Validated using CAN bus-compatible data, the framework achieves exceptional metrics, including an RMSE of 0.3921, an MAE of 0.0557, an R 2 of 0.9999, and a SMAPE of 0.84%, setting a new standard in the field. The hybrid model surpasses traditional approaches by aligning simulated outputs with real-world dynamics, offering scalability across varying conditions and demonstrating significant potential for real-time predictive maintenance, operational optimization, and integration with digital twin technologies.

This paper presents novel observations on the impact of ground-bounce noise (GBN) on logic-HIGH levels of the output response of Complementary Metal Oxide Semiconductor (CMOS) inverter circuits. While the impact of power supply noise on logic-HIGH is undeniable and comprehensively studied in the literature, the impact of ground bounce is perceived, and the analysis is majorly focused on logic-LOW. This work demonstrates as well as analyses the significant impact of GBN on logic-HIGH and the observations are supported by a theoretical investigation using both the analytical and small-signal approaches. Several examples are considered to validate the novel observations considering both the simulation and measurement-based case studies. The simulation-based case studies are performed using several CMOS technologies, and measurement-based case studies are performed using standard HEX inverter Integrated Circuits (ICs). There is a close correlation observed between the practical observations and the corresponding theoretical foundation.

This research paper explores the application of advanced Natural Language Processing (NLP) techniques to enhance the realism and immersion of player-NPC interactions in Augmented Reality (AR) and Virtual Reality (VR) games. We propose novel approaches to improve existing NLP models, focusing on context-awareness, emotional intelligence, and realtime adaptation. Our findings suggest that these enhancements significantly improve the naturalness and depth of conversations with AI-driven NPCs, leading to more engaging and immersive gaming experiences in AR/VR environments.

The emergence of 6G networks promises to revolutionise telecommunications by incorporating end-to-end intelligence for improved performance, driven by new radio and other access technologies. On the other hand, optical technologies are crucial for enabling long-distance integration among various domains and network technologies. This integration is vital for future networks to meet Key Performance Indicators (KPIs) like bitrate and latency, and Key Value Indicators (KVIs) such as reliability and privacy. The envisioned end-to-end intelligence can drive optical networks to consistently provide highquality connectivity and handle dynamic computing and network demands in 6G. This paper explores end-to-end intelligence within 6G networks, emphasising its potential to optimise service delivery. We envision a 6G-ready architecture for delivering end-to-end services across multiple domains interconnected by optical technologies. This architecture uses Artificial Intelligence and Machine Learning to support lifecycle management of infrastructures and network services. It includes an Intelligent Optical Cross-Domain Orchestrator and Optical Controllers at the Transport and Metro domains, as well as Intra-Domain and Cross-Domain Service Orchestration, efficiently allocating optimum resources to services ensuring robust and seamless performance in future networks.

This research paper explores the cutting-edge domain of AI-driven procedural content generation (PCG) for virtual reality (VR) and augmented reality (AR) environments. We investigate the synergy between artificial intelligence techniques and PCG methodologies to create dynamic, adaptive, and immersive content for VR/AR applications. The paper presents novel algorithms, frameworks, and case studies that demonstrate the potential of AI-PCG in enhancing user experiences, reducing development costs, and enabling personalized content creation at scale. We also address the unique challenges posed by VR/AR environments, such as real-time performance requirements, spatial coherence, and user interaction complexities. Our findings suggest that AI-driven PCG has the potential to revolutionize content creation for VR/AR, opening new avenues for interactive storytelling, adaptive game design, and immersive training simulations.

This paper introduces DynamicRenderNet, a novel neural network architecture designed to optimize graphics rendering in real-time Augmented Reality (AR) and Virtual Reality (VR) games. By adaptively adjusting rendering parameters based on scene complexity and user interaction, DynamicRenderNet significantly improves frame rates and reduces latency. Our model utilizes a deep learning approach to analyze real-time input data, such as camera feed, user head movements, and scene geometry, to predict optimal rendering settings. This enables dynamic resource allocation and prioritization, leading to a more efficient and responsive rendering pipeline. Experimental results demonstrate substantial improvements in frame rates and reduced latency, resulting in a more immersive and visually appealing AR/VR experience.