Recently, 3D occupancy prediction, a camera-only perception task, has garnered significant attention for addressing key limitations in traditional 3D object detection, such as overlooking uncommon categories and failing to capture complex geometric shapes. However, current methods for occupancy prediction face two major challenges. First, they require high computational resources, making deployment on non-GPU devices impractical. Second, height estimation is often inaccurate due to the overlooked imbalance in height distribution within datasets. To address these limitations, we propose two innovative strategies. First, we eliminate the use of computationally expensive components such as transformer operators, depth estimation modules, and 3D convolutions. Second, we introduce a focused weighting mechanism to improve height-related accuracy. Building on these strategies, we introduce ConvOcc, an efficient and deployment-friendly framework composed entirely of 2D convolutions. ConvOcc features: (1) Feature Fuse Module for enhanced multi-scale 2D feature fusion, (2) Voxel-to-Image View Transformation for rapid conversion of 2D image features to 3D voxel space, (3) Squash and Stretch Module to simplify complex 3D voxel computations into a more efficient 2D BEV form, (4) Height-Attention Multi-Scale BEV Fusion Module for dynamic reweighting of BEV features based on height, and (5) Multi-Frame Temporal Fusion Strategy for denser voxel feature extraction. Extensive ablation studies validate the effectiveness and efficiency of our approach. ConvOcc achieves 2× FPS with a mean IoU of 36.1 on the Occ3D-nuScenes dataset, all while maintaining deployment-friendly requirements. This work challenges the conventional reliance on 3D methods for occupancy prediction, demonstrating that it can be effectively and efficiently addressed using a 2D-based approach.

Quantum networks hold the potential to revolutionize a variety of fields by surpassing the capabilities of their classical counterparts. Many of these applications necessitate the sharing of high-fidelity entangled pairs among communicating parties. However, the inherent nature of entanglement leads to an exponential decrease in fidelity as the distance between quantum nodes increases. This phenomenon makes it challenging to generate high-fidelity entangled pairs and preserve information in quantum networks. To tackle this problem, we utilized two strategies to ensure high-fidelity entangled pairs and information preservation within a quantum network. First, we use closeness centrality as a metric to identify the closest nodes in the network. Second, we introduced the trace-distance based path purification (TDPP) algorithm, specifically designed to enable information preservation and path purification entanglement routing. This algorithm identifies the shortest path within quantum networks using closeness centrality and integrates trace-distance computations for distinguishing quantum states and maintaining end-toend (E2E) entanglement fidelity. Simulation results demonstrate that the proposed algorithm improves network throughput and E2E fidelity while preserving information compared to existing methods.

This paper considers the optimal control problem of steering a Dubins vehicle with a decaying speed that depends linearly on the turn-rate to a terminal point and heading with maximum kinetic energy. Pontryagin’s minimum principle is applied and the extremals are identified as sequences of straight segments and spiral-shaped turns described by involutes of a circle. A method is proposed to exactly transcribe the optimal control problem into a series of finite dimensional optimizations. Each of these optimizations determines a locally optimal candidate path that reaches the terminal state. The lowest cost path among the candidate paths is then identified. The approach is demonstrated through several numerical examples.

This paper presents two units of Y-band (325-500 GHz) and two units of D-band (110-170 GHz) all-pole inline Chebyshev band-pass filters (BPFs) using high-precision computer numerically controlled milling. The filters are based on six-coupled resonant cavities and are designed for a 20 dB return loss in a 10 GHz bandwidth centered around 470 GHz and 160 GHz, respectively. The filter sections are designed using the standard waveguides of the next frequency band to guarantee single-mode propagation, and taper sections are placed for the transition. Statistical analyses of the design parameters are performed to estimate the manufacturing tolerance. Microscope measurements are done and it is seen that machine tolerance on cavity lengths of about 3.5 microns causes a 2 GHz center frequency shift in Y-band. Measurements validate the analyses. The fabrication repeatability in the D-band is more consistent than in the Y-band. Indeed, two Y-band BPF units have a center frequency shift difference of 6 GHz while the the D-band BPF units have almost identical responses.

Over the past decade, there has been growing interest in using human behavioral and physiological data to detect Social Anxiety Disorder (SAD). Machine learning and deep learning techniques that use multimodal sensing have emerged as promising tools for detecting SAD characteristics. Additionally, extensive research on technology-assisted psychological interventions for SAD aims to enhance treatment efficacy and address the shortcomings of existing treatments by exploring how these interventions can be tailored to individual anxiety levels, symptom severity, and personal preferences. This review provides an overview of approaches for generalised SAD, covering advancements in both sensing and interventions while highlighting the potential of affective computing. It synthesises key insights on current emerging trends, identifies research gaps, and outlines directions for future research.

This article presents a novel design methodology for bandpass filters applied to rectangular waveguide (RW) technology using TE 201 singlets with inductive irises. The conventional circuit model for an inductive iris interconnecting two RW sections, which typically handles only the TE 10 mode, is extended to include the propagation of the TE 20 mode through a wider RW section. The impact of the reactances at the source and load terminals of the circuit model is analyzed to develop a unique solution for the synthesis of the singlet coupling matrix. The proposed methodology improves design flexibility, allowing for filters with both cascaded singlet and cascaded triplet topologies. This is demonstrated through two illustrative designs. The first design employs propagating RW sections as phase shifters to interconnect TE 201 singlets, which is the most common approach. The second design, however, replaces these sections with TE 101 resonant cavities. This substitution offers advantages in spuriousfree range, volume, and mass. Both prototypes were successfully manufactured and measured to validate the approach experimentally. In addition, the proposed method simplifies electromagnetic simulations by focusing on the irises at a single frequency point, thus avoiding the need to analyze entire singlet structures over full frequency sweeps.

Camouflage is an adaptive strategy that allows organisms and objects to blend with their surrounding environments, thereby evading detection by predators or adversaries. Current artificial methods, including manual painting, computer-aided techniques, and deep learning approaches, face significant challenges related to manual intervention, scalability, and generalization across diverse and non-calibrated scenes. In this paper, we propose a novel few-shot diffusion-based method for camouflage pattern generation, called Camouflage Diffusion (CamoDiff). Our method consists of two distinct stages: meta learning and few-shot learning. During the meta learning stage, our method employs a diffusion-based architecture to automatically generate camouflage patterns from noise, thus it eliminates manual intervention and enables scalable production. Additionally, our approach integrates the guidance mean absolute error (GMAE) loss in the few-shot learning stage. This allows the generated camouflage to adapt seamlessly to new environments with minimal retraining, regardless of varying viewpoints. We also introduce a comprehensive camouflage dataset and generation tool, which can be used as benchmarks for future research. Experimental results demonstrate that CamoDiff outperforms existing state-of-the-art methods in camouflage pattern generation on different datasets and metrics.

Cross-topology generalization is essential for improving the practicality of deep learning models in power system analysis (PSA) tasks. While large pre-trained models hold substantial promise for PSA, existing research lacks insight into which neurons are pivotal for topology generalization, leading to inefficient fine-tuning processes. This letter reveals the adaptation mechanism of deep PSA models to topological changes by introducing a causally-guided framework to analyze functional differences among neurons. A subset of neurons, termed topology-sensitive neurons, is identified as selectively activated by specific topological inputs and shown to causally influence model performance. Using integrated gradient methods, these neurons are localized and targeted with a masked low-rank adaptation (LoRA) strategy, which prioritizes parameter updates based on causal impact rather than indiscriminate fine-tuning. Experimental results demonstrate that focusing on topologysensitive neurons improves fine-tuning efficiency, reducing trainable parameters by 89% while maintaining high accuracy.

Soliton crystal microcombs, as a new type of Kerr frequency comb, offer advantages such as higher energy conversion efficiency and a simpler generation mechanism compared to traditional soliton microcombs. They have a wide range of applications in fields like microwave photonics, ultra-high-speed optical communication, and photonic neural networks. In this review, we discuss recent developments of soliton crystals microcombs and analyze the advantages and disadvantages of generating soliton crystal microcombs utilizing different mechanisms. First, we briefly introduce the numerical model of optical frequency combs. Then, we introduce the generation schemes for soliton crystal microcombs based on various mechanisms, such as utilizing an avoided mode crossing, harmonic modulation, bi-chromatic pumping, and the use of saturable absorbers. Finally, we discuss the progress of research on soliton crystal microcombs in the fields of microwave photonics, optical communication, and photonic neural networks. We also discuss the challenges and perspectives of soliton crystal microcombs.

Federated Learning (FL) is a decentralized ML approach that can be used for intrusion detection in Internet of Things (IoT) devices. It involves the local training of AI models and their aggregation at a central server. This methodology eliminates the need for data sharing between IoT devices while fostering collaborative model enhancement. Nonetheless, concerns arise due to the lack of transparency surrounding the shared local models and the aggregation techniques employed. This lack of transparency can potentially lead to model poisoning attacks and hinder collaborators from using alternative aggregation methods that better align with their specific use cases. To address this issue, this paper proposes a blockchainbased approach with FL to ensure transparent and immutable records of model updates, thereby bolstering security and trust for intrusion detection in IoT devices. In contrast to traditional synchronization or periodic update-based approaches, this paper proposes a novel time-independent aggregation method in FL blockchain, allowing for flexibility. Additionally, the proposed blockchain allows various users to utilize their own aggregation methods rather than a fixed one, based on their needs, resources, and availability. We also develop a user interface for the proposed blockchain system that assists in visualizing different aspects of the method, such as model aggregation. The proposed system is tested using traditional metrics like AI model performance as well as extensive user testing.

We describe a security constraint evaluation (SCE) Python package developed in the SCY 0 project. The SCE package is a simplified version of the simultaneous feasibility test (SFT) developed in the HIPPO project on security constrained unit commitment (SCUC). The HIPPO SFT achieved speedups over a typical commercial SCE product used in large scale electricity markets by using a mathematical technique based on the Sherman-Morrison-Woodbury (SMW) formula for the inverse of a matrix with a low rank update to handle contingencies where some branches are opened or closed. We use the SCE package to study the computational kernels of the SMW approach to understand and improve on the performance of the HIPPO SFT. We describe novel methods for these kernels using Python acceleration techniques including just in time (JIT) compilation and graphical processing unit (GPU) deployment. Run time and memory performance results for these methods on large scale test problems demonstrate scaling properties. Some of these methods achieve speedup factors of 30 to 80, relative to a baseline implementation and to the fastest HIPPO method. After the HIPPO results, these results further support the use of the SMW approach, and they show acceleration techniques can improve SCUC solver performance.

This paper presents a framework for training a Gaussian process (GP) to estimate a steady urban wind field from a sparse set of wind measurements by leveraging training data collected from computational fluid dynamics (CFD) simulations. Gaussian process models for spatial estimation often use measurement locations as the input space with proximity-based covariance functions. This work investigates including building morphology features into the GP model that are defined by the signed distance field (SDF) and its gradient evaluated at a pattern of points around each sample location. Augmenting the measurement locations with different subsets of building morphology features leads to unique feature spaces. Several different GP models are trained using various feature spaces and covariance functions, including with a coregion covariance function that allows simultaneous training over multiple CFD datasets for different urban geometries. A framework is developed to generate CFD wind field data for a set of randomized geometries, build various feature spaces, and perform the estimation with the proposed GP models. The framework is evaluated with a simple environment that consists of two buildings with randomized position and geometry in a wind field with constant inflow magnitude and direction. Results are presented comparing the estimation performance across different GP models with an increasing number of optimization iterations. The computation versus accuracy trade-off of using hyperparameters trained over multiple similar prior CFD datasets, rather than hyperparameters that are optimized on-the-fly, over a single dataset is also demonstrated.

This paper presents a dynamic model of a quadrotor UAV affected by a blast wave that impinges on the vehicle. The body of a quadrotor is decomposed into four spherical motor elements connected by rods to a central spherical body. Existing models for the pressure and induced air velocity of a blast wave are adapted for this work and compared to computational fluid dynamics (CFD) simulations. The response of a single sphere to a blast wave is characterized by integrating the pressure over the sphere to obtain an impulse due to the pressure wave and drag. This analysis is extended to the multi-sphere model of the quadrotor UAV. As the blast interacts with the UAV, the non-uniform pressure over the body results in forces and moments that perturb the vehicle. Similarly, the blast-induced wind generates drag forces that are superimposed. The response of the vehicle to the blast pressure and wind is investigated by numerically integrating the equations of motion over the short duration of the blast to determine the terminal state of the UAV after 100 ms when the surroundings return to ambient conditions.

According to the World Health Organization [1], cancer is the leading cause of death globally, with nearly 10 million deaths reported in 2020. Each year, around 400,000 children are diagnosed with cancer. The most common types of cancer varying by region.Laryngeal cancer is a type of head and neck cancer. According to Cancer.net, over 184,000 cases of laryngeal cancer were diagnosed worldwide in 2020 [3]. While it may not be as prevalent as breast, lung, or prostate cancer, laryngeal cancer presents unique opportunities for early diagnosis using audio and image data. Approximately 60% of laryngeal cancers are found in the glottis, and about 35% are in the supraglottis. For cancers confined to the larynx, the five-year survival rates are 84% for glottis cancers and 61% for supraglottis cancers [3]. Figure 1, provided by the American Society of Clinical Oncology, illustrates the medical anatomy of the glottis and supraglottis in the human throat.

This paper provides a short overview of how artificial intelligence (AI) can be used on voice data for early detection of various medical conditions, thus accelerating triaging and improving patient outcome.Clinical Relevance — Worldwide more than 10 million people suffer from Parkinson’s disease, 5% of adults (~280 million) suffer from depression and more than 184000 laryngeal cancer diagnosis were made in 2020. All these diseases and many more can be potentially detected in earlier phases by using AI on voice, hence giving medical professionals an AI-augmented way to address such conditions in early stages.

The adoption of Electric Vehicles (EVs) is a transformative step towards reducing carbon dioxide (CO2) emissions and achieving sustainability, aligning with the goals of IMT-2030. However, this transition spans multiple domains, including the Smart Grid (SG), Transportation Systems (TS), and Information and Communication Technology (ICT), each with its own resource limitations. For example, the ICT sector faces constraints in spectrum, transmit power, and computational capacity, which are further strained as the number of EVs increases. Addressing these challenges requires a holistic approach to optimize resource utilization, enhance Energy Efficiency (EE), and manage the complex interactions across these interconnected sectors. This survey aims to identify the critical resources required for EV adoption and examine the challenges arising from their limitations across SG, TS, and ICT. It explores potential solutions and Key-Enabling Technologies (KETs) while reviewing cross-sector resource allocation frameworks to address these challenges. Finally, the survey highlights open problems and future research directions, providing a comprehensive foundation for advancing EV adoption.

This research enhances and evaluates the performance of a Decentralized Nash Bargaining (DNB) adaptive traffic signal controller that operates a flexible National Electrical Manufacturers Association (NEMA) phasing and timing scheme responding dynamically to the fluctuating traffic demand. The DNB controller is enhanced to 1) use traffic density estimates instead of queues to optimize signal timings; 2) to consider the eight-phase two-ring NEMA controller configuration within the game-theoretic approach; and 3) to consider dynamically adaptable control time steps. The enhanced DNB controller is benchmarked against 1) fixed-time traffic signal control using the state-of-the-practice Webster's method and an emerging Laguna-Du-Rakha (LDR) method for computing the optimum cycle length; 2) state-of-practice actuated traffic signal control; and 3) state-of-art Reinforcement Learning (RL) traffic signal control. The controller is tested on two isolated signalized intersections demonstrating enhanced overall intersection performance compared to the baseline pre-timed and actuated controllers at various demand levels and offers comparable performance to a previously developed RL controller. Specifically, the DNB controller results in a decrease in the average vehicle delay and queue size by up to 54% and 63%, respectively compared to Webster's state-of-the-practice pre-timed control. Unlike the RL controller, the DNB controller requires no pre-training while adapting to fluctuating traffic conditions, thereby providing a flexible framework for reducing traffic congestion at signalized intersections. As such, this research contributes to the development of smarter and more responsive urban traffic control systems.

This survey paper provides an extensive review of the application of Large Language Models (LLMs) in network operations and management (NO&M). It outlines the transformation in NO&M driven by LLMs, emphasizing their potential to address challenges such as network design, automation, optimization, and security. The paper explores how LLMs can enhance traditional methods by automating complex tasks, improving network agility, and providing scalable and adaptive solutions to emerging network demands. Key contributions include a detailed discussion on LLM-enabled techniques that address issues like real-time adaptability, scalability, security, and intent-based management. The survey categorizes existing research into several key areas and identifies current limitations, such as integration with legacy systems, explainability, data privacy, and scalability. Moreover, it highlights future research directions, including the need for scalable architectures, energy-efficient models, enhanced security protocols, and ethical considerations in deploying AI-driven solutions. The findings are expected to drive innovation and provide valuable insights for researchers and practitioners aiming to leverage LLMs for advanced networking tasks.