School of Engineering, Computer and Mathematical Sciences - Te Kura Mātai Pūhanga, Rorohiko, Pāngarau

Permanent link for this collectionhttps://hdl.handle.net/10292/553

AUT is home to a number of renowned research institutes in engineering, and computer and mathematical sciences. The School of Engineering, Computer and Mathematical Sciences strong industry partnerships and the unique combination of engineering, computer and mathematical sciences within one school stimulates interdisciplinary research beyond traditional boundaries. Current research interests include:

Artificial Intelligence; Astronomy and Space Research;

Biomedical Technologies;

Computer Engineering; Computer Vision; Construction Management;

Data Science;

Health Informatics and eHealth;

Industrial Optimisation, Modelling & Control;

Information Security;

Mathematical Sciences Research; Materials & Manufacturing Technologies;

Networking, Instrumentation and Telecommunications;

Parallel and Distributed Systems; Power and Energy Engineering;

Software Engineering; Signal Processing; STEM Education;

Wireless Engineering;

Browse

Now showing 1 - 20 of 1792

Joint Effect of Signal Strength, Bitrate, and Topology on Video Playback Delays of 802.11ax Gigabit Wi-Fi
(MDPI AG, 2026-01-26) Sarkar, Nurul; Gul, Sonia
This paper presents a performance evaluation of IEEE 802.11ax (Wi-Fi 6) networks using a combination of real-world testbed measurements and simulation-based analysis. The paper investigates the combined effect of received signal strength (RSSI), application bitrate, and network topology on video playback delays of 802.11ax. The effect of frequency band and client density on system performance is also investigated. Testbed measurements and field experiments were conducted in indoor environments using dual-band (2.4 GHz and 5 GHz) ad hoc and infrastructure network configurations. OMNeT++ based simulations are conducted to explore scalability by increasing the number of wireless clients. The results obtained show that the infrastructure-based deployments provide more stable video playback than the ad hoc network, particularly under varying RSSI conditions. While the 5 GHz band delivers higher throughput at a short range, the 2.4 GHz band offers improved coverage at reduced system performance. The simulation results further demonstrate significant degradation in throughput and latency as client density increases. To contextualize the observed performance, a baseline comparison with 802.11ac is incorporated, highlighting the relative improvements and remaining limitations of 802.11ax within the evaluated signal and load conditions. The findings provide practical deployment insights for video-centric wireless networks and inform the optimization of next-generation Wi-Fi deployments.
Hierarchical Switch Fault Diagnosis Based on Transformer Algorithm in Four-leg Inverters of Stand-alone Wind Energy Conversion Systems
(Elsevier, 2026-01-27) Heidari, Jalqal; Peykarporsan, Rasool; Oshnoei, Soroush; Lie, Tek Tjing; Vandevelde, Lieven; Crevecoeur, Guillaume
With the increasing development of renewable energy resources, stand-alone structures are gaining more attention. Among these, wind energy systems are particularly notable because of their advantages, including sustainability, low operational expenses, and minimal environmental impact. Due to the challenges of load balancing in such systems, four-leg inverters have emerged as a viable solution, offering improved performance under unbalanced load conditions. However, like all inverters, they remain susceptible to internal faults. Accordingly, this paper proposes a hierarchical two-level Transformer-based model to detect switch internal faults, including open-circuit and short-circuit in four-leg inverters. The OPAL-RT hardware-in-the-loop setup was used to generate data in various scenarios to validate the efficiency of the proposed framework. The results demonstrate that the developed technique can effectively classify fault types and identify faulty switches compared to state-of-the-art algorithms and single-level structures.
A Comparative Conceptual Analysis of CO₂ Heat Pump Dryers With Closed-loop and Open-loop Air Cycles
(Elsevier, 2026-01-25) Wang, Jay
This study has comprehensively compared and analysed CO₂ heat pump dryers operating under closed-loop and open-loop air cycles to evaluate their energy efficiency and drying performance. Unlike the conventional closed-loop air cycle that uses dry recirculated air as its inlet, the open-loop air cycle operates only with fresh ambient air. The physical models and working principles have been illustrated using psychrometric charts, and the influence of moisture variation has been considered in the fin-and-tube heat exchanger design for both the gas cooler and the evaporator. In the case study under typical hot and humid climate conditions (ambient temperature of 40 °C), the simulation compares three cycles over an air mass flow rate ranging from 0.5 kg/s to 1 kg/s. The open-loop air cycle with a wet air outlet achieves the largest heating capacity, i.e.: 17.44 kW at 1 kg/s, because the air is cooled in the evaporator first, allowing a greater temperature rise in the gas cooler. The open-loop air cycle with a dry air outlet produces the highest air temperature after the gas cooler, i.e.: 60.8 °C at 0.5 kg/s, which increases the air’s moisture absorption capacity. Compared with the closed-loop air cycle, the open-loop air cycle with dry air outlet proves more efficient for drying, delivering a shorter drying time (27.77 min at 0.5 kg/s) and a higher drying efficiency (0.8640 kg/kWh at 0.5 kg/s). Although the open-loop air cycle with a wet air outlet achieves the highest coefficient of performance of 2.31 at 1 kg/s, its drying performance declines obviously at higher mass flow rates, with specific moisture extraction rate dropping to 0.0767 kg/kWh. Overall, the configuration of open-loop air cycle with dry air outlet is the superior option, as it combines the shortest drying time and the highest specific moisture extraction rate, which are two critical metrics for heat pump dryers.
An Attention-based BERT–CNN–BiLSTM Model for Depression Detection From Emojis in Social Media Text
(MDPI AG, 2025-12-03) Thekkekara, Joel Philip; Yongchareon, Sira
Depression represents a critical global mental health challenge, with social media offering unprecedented opportunities for early detection through computational analysis. We propose a novel BERT–CNN–BiLSTM architecture with attention mechanisms that systematically integrate emoji usage patterns—fundamental components of digital emotional expression overlooked by existing approaches. Evaluated on the SuicidEmoji dataset, our model achieves 97.12% accuracy, 94.56% precision, 93.44% F1-score, 85.67% MCC, and 91.23% AUC-ROC. Analysis reveals distinct emoji patterns: depressed users favour negative emojis (😔 13.9%, 😢 12.8%, 💔 6.7%) while controls prefer positive expressions (😂 16.5%, 😊 11.0%, 😎 10.2%). The attention mechanism identifies key linguistic markers, including emotional indicators, personal pronouns, and emoji features, providing interpretable insights into depression-related language. Our findings suggest that the integration of emojis substantially improves optimal social media-based mental health detection systems.
A Hierarchical Federated Continual Learning Framework for Dynamic and Heterogeneous IoV
(Institute of Electrical and Electronics Engineers (IEEE), 2026-01-15) Chen, Yiming; Wu, Celimuge; Zhong, Lei; Lin, Yangfei; Du, Zhaoyang; Bao, Wugedele; Chong, Peter Han Joo
Traditional federated learning (FL) architectures face challenges in handling heterogeneous data and dynamic tasks, often resulting in catastrophic forgetting when new training tasks are continuously introduced. Federated continual learning (FCL) integrates the privacy-preserving capabilities of FL with the knowledge retention and incremental update mechanisms of continual learning, effectively mitigating catastrophic forgetting and protecting user privacy. However, existing FCL solutions largely overlook the unique requirements of Internet of Vehicles (IoV) scenarios, such as data heterogeneity and dynamic task management. To address these challenges, we propose a novel framework, hierarchical federated continual learning (Hier-FCL), which incorporates local continual learning via optimized experience replay and meta-knowledge distillation, along with dynamic client clustering to tackle data heterogeneity. Additionally, a hierarchical aggregation mechanism is employed to enhance scalability and adaptability in diverse IoV scenarios. Experiments conducted in mixed-task environments using multiple datasets demonstrate that Hier-FCL outperforms baseline algorithms in terms of retained accuracy and backward transfer impact, validating its effectiveness in mitigating catastrophic forgetting and managing heterogeneous client data.
Beyond the Generalist
(IGI Global, 2026-01-19) Thorpe, Stephen J; Krstić, Livia
International frameworks and accreditations define the core competencies required of information technology (IT) project managers. Among these, technical skills are often cited as important, particularly in IT-focused projects. However, the technical competencies required—and the extent to which project managers should possess them—remain unclear. The literature on this topic is limited, though existing studies indicate that technical proficiency contributes to project success in technical domains. To explore this gap, semi-structured interviews with IT project managers and project participants were undertaken to examine perceptions of technical skills. Findings reveal a divide between participants with technical education, who emphasized the necessity of technical expertise, whereas those without technical qualifications highlighted communication, motivation, and attitude as most critical. The study contributes insights into the strategic value that technical capability adds to IT project management effectiveness through the strategic capability model for technical project management.
Automated Food Weight and Content Estimation Using Computer Vision and AI Algorithms: Phase 2
(MDPI AG, 2025-12-22) Gonzalez, Bryan; Garcia, Gonzalo; Velastin, Sergio A; GholamHosseini, Hamid; Tejeda, Lino; Ramirez, Heilym; Farias, Gonzalo
The work aims to leverage computer vision and artificial intelligence technologies to quantify key components in food catering services. Specifically, it focuses on content identification and portion size estimation in a dining hall setting, typical of corporate and educational settings. An RGB camera is employed to capture the tray delivery process in a self-service restaurant, providing test images for content identification algorithm comparison, using standard evaluation metrics. The approach utilizes the YOLO architecture, a widely recognized deep learning model for object detection and computer vision. The model is trained on labeled image data, and its performance is assessed using a precision-recall curve at a confidence threshold of 0.5, achieving a mean Average Precision (mAP) of 0.873, indicating robust overall performance. The weight estimation procedure combines computer vision techniques to measure food volume using both RGB and depth cameras. Subsequently, density models specific to each food type are applied to estimate the detected food weight. The estimation model's parameters are calibrated through experiments that generate volume-to-weight conversion tables for different food items. Validation of the system was conducted using rice and chicken, yielding error margins of 5.07% and 3.75%, respectively, demonstrating the feasibility and accuracy of the proposed method.
Advancing Physical Activity Monitoring Through Bioimpedance Measurement: A Review
(IOP Publishing, 2026-01-09) Jacobs, Ifeanyi; Lowe, Andrew; Garcia, Lorenzo; Zhang, Huiyang
Bioimpedance measurements have gained significant attention due to their ability to assess body composition, muscle health, and internal physiological states without the need for intrusive procedures. This review paper explores the advancements and applications of bioimpedance technology, a non-invasive and cost-effective method for real-time monitoring of physiological parameters and physical activities. It discusses key measurement modalities such as bioelectrical impedance analysis (BIA), electrical impedance myography (EIM), and electrical impedance tomography (EIT), highlighting their unique advantages and applications. It also examines the role of biopotential electrodes, both polarizable and non-polarizable, in ensuring accurate physiological measurements. Despite challenges such as low spatial resolution, motion artifacts and sensitivity to electrode placement, the review highlights promising solutions. These include the integration of hybrid sensor systems, machine learning algorithms for signal interpretation, and the development of wearable and flexible electronics. The paper concludes by emphasizing the growing potential of bioimpedance technology in fields such as sports science, rehabilitation, personalized healthcare, fitness monitoring, and human-machine interaction, suggesting a future where continuous physiological monitoring becomes seamlessly embedded in daily life.
Modeling and Performance Assessment of a NeWater System Based on Direct Evaporation and Refrigeration Cycle
(MDPI AG, 2026-01-17) Huo, Yilin; Hu, Eric; Wang, Jay
At present, the global shortage of water resources has led to serious challenges, and traditional water production technologies such as seawater desalination and atmospheric water harvesting have certain limitations due to inflexible operation and environmental conditions. This study proposes a novel water production system (called “NeWater” system in this paper), which combines saline water desalination with atmospheric water-harvesting technologies to simultaneously produce freshwater from brackish water or seawater and ambient air. To evaluate its performance, an integrated thermodynamic and mathematical model of the system was developed and validated. The NeWater system consists of a vapor compression refrigeration unit (VRU), a direct evaporation unit (DEU), up to four heat exchangers, some valves, and auxiliary components. The system can be applied to areas and scenarios where traditional desalination technologies, like reverse osmosis and thermal-based desalination, are not feasible. By switching between different operating modes, the system can adapt to varying environmental humidity and temperature conditions to maximize its freshwater productivity. Based on the principles of mass and energy conservation, a performance simulation model of the NeWater system was developed, with which the impacts of some key design and operation parameters on system performance were studied in this paper. The results show that the performances of the VRU and DEU had a significant influence on system performance in terms of freshwater production and specific energy consumption. Under optimal conditions, the total freshwater yield could be increased by up to 1.9 times, while the specific energy consumption was reduced by up to 48%. The proposed system provides a sustainable and scalable water production solution for water-scarce regions. Optimization of the NeWater system and the selection of VRUs are beyond the scope of this paper and will be the focus of future research.
A Robust Secure and Energy-Aware Cross-Layer Framework for IoT Networks
(IEEE, 2026-01-14) Mustafa, Rashid; Sarkar, Nurul I; Mohaghegh, Mahsa; Pervez, Shahbaz
The dual challenges of energy constraints and multi-layered cyber threats must be addressed in order to secure Internet of Things (IoT) environments. To overcome the above problems, we propose a secure and energy-aware cross layer framework for IoT networks. Our framework is based on the combined role-based access control, machine learning based anomaly detection, and lightweight encryption. We explore context-aware defenses that can remain scalable and energy efficient while dynamically adapting to changing attack vectors. The performance of the proposed framework is evaluated using real hardware (Z1 and EXP430F5438 motes) after being validated by simulations on the Cooja and NS-3 platforms. The results demonstrate up to 30% energy savings over AES while preserving high detection performance for both active and passive threat models and over 95% packet delivery. These results highlight the necessity of adaptive, multi-layer strategies for contemporary IoT deployments and show that a secure, scalable, and energy conscious IoT design is feasible.
H∞ Bipartite Synchronization Composite Antidisturbance Control of Hidden Markov Jump Reaction–Diffusion Neural Networks
(Institute of Electrical and Electronics Engineers (IEEE), 2025-12-30) Wang, X; Sun, L; Wang, YL; Lie, T
This article investigates the problem of composite H∞ control for cooperation–competition networks with hidden Markov jump parameters reaction–diffusions dynamics. Considering the difficulty of directly obtaining the mode information of systems, a continuous-time hidden Markov jump model is employed to represent the joint jump process. Specifically, the hidden process stands for the dynamics of real systems, which cannot be precisely known but can be observed through a detector. Due to the existence of multiple disturbances, the performance of the aforementioned systems can be deteriorated. To reduce the influence of these disturbances, a composite disturbance observer-based controller is constructed, which combines a disturbance observer with a feedback control mechanism. This design significantly improves the robustness and antidisturbance capability of systems. Then, sufficient criteria are derived to guarantee that the bipartite synchronization error system (BSES) is stochastically stable and meets a desired performance index. Finally, the effectiveness of the proposed control method is verified through the performance analysis.
Mitigation of Cyber-Physical Attacks in Industry 4.0 Using Secure Function Blocks
(ACM, 2025-12-29) Wu, Steph; Allen, Nathan; Baird, Alex; Pearce, Hammond; Roop, Partha
As Industry 4.0 drives the Fourth Industrial Revolution, Cyber-Physical Systems (CPSs) have become central to industrial automation. These systems integrate software with physical processes, significantly improving the efficiency and adaptability. However, this integration also expands the attack surface, exposing systems to Cyber-Physical-attacks (CP-attacks) that can target either the computational components, physical devices, or both. The impact of such attacks can be catastrophic, ranging from system disruption to physical damage. Although numerous techniques have been developed to detect and mitigate these threats, industrial standards are often not incorporated into the design of these methods. This limits their deployment within the Industry 4.0 systems, where standard compliance is critical. To this end, we extend IEC 61499, an emerging standard being considered in Industry 4.0, that uses reusable artefacts called function blocks. We formalise the mitigation of CP-attacks using a novel method based on Bi-directional Runtime Enforcement (Bi-RE) using a standards compliant approach called Secure Function Blocks (SFBs). This approach automatically generates the necessary enforcers from a timed specification language called Valued Discrete Timed Automaton (VDTA). We illustrate our approach using the case study of a water treatment system, and highlight the low overhead associated with it. This approach allows for runtime techniques to be applied with minimal changes to many Industry 4.0 applications.
Optimising the Scheduling of System Level Logical Execution Time Systems
(ACM, 2025-12-29) Lee, Jamie; Allen, Nathan; Kuo, Matthew MY; Yip, Eugene
The paradigm of Logical Execution Time (LET) tasks is widely adopted by major tool vendors for designing deterministic and time-predictable software in multi-core systems, particularly in the automotive industry. To extend the use of LET in distributed environments, System Level Logical Execution Time (SL-LET) has been developed to effectively manage communication and delays between networked devices. However, there is currently a lack of open-source tools available for SL-LET, and the task allocation and scheduling problem for SL-LET remains unsolved. To address these concerns, we introduces a novel Integer Linear Programming (ILP)-based optimisation approach for SL-LET task allocation and scheduling, focusing on minimising core utilisation and average system response times. To illustrate the effectiveness of the approach, we benchmark our ILP-based solution against a traditional core allocation heuristic across multiple task sets. Through this evaluation, our approach, when compared to the heuristic, is able to demonstrate average response times that are 26.9% smaller.
Tuning Into My Heart Through Wearables: Towards a Formal Cardiac Digital Twin
(ACM, 2025-12-29) Roop, Partha; Allen, Nathan; Kazemi, Shahab
Digital Twins (DTs) mimic a physical system using a digital version of the real system. While these have been explored in many domains, digital twins of human organs are yet to be created, especially those that are inspired by formal methods. To this end, we propose the first Cardiac Digital Twins (CDTs) by leveraging two key innovations from our research group. The first is a real-time model of the heart, that is based on a network of hybrid automata to represent the cardiac conduction system that mimics the rhythmic electrical activity of a normal heart. The model can be parametrised to exhibit disease states in real-time and this approach is being used by MathWorks for closed-loop validation of pacemakers in real-time. This work has raised the interest of both device manufacturers and certification agencies, especially in the USA. Our group has expertise in digital biomarkers obtained from wearables, such as Electrocardiograms (ECGs) and Photoplethysmograms (PPGs). These provide a window into the cardiac cycle and we have already shown that the two signals are strongly correlated. Hence, a second innovation is related to using wearables to personalise the real-time heart model, so that the model generates ECGs matching that of an individual in different states. Our approach paves the way for developing personalised therapies, real-time monitoring, and accurate estimation of heart rate variability.
Combine Meta-Learning with Feature Alignment for Cross-Domain Heterogeneous Hyperspectral Image Classification
(Institute of Electrical and Electronics Engineers (IEEE), 2026-01-12) Ye, Minchao; Jin, Yuheng; Zhao, Jianwei; Yan, Weiqi; Qian, Yuntao
The scarcity of labeled samples results in the challenge of small-sample-size in hyperspectral image (HSI) classification. Transfer learning offers hope for solving this problem. In cross-domain transfer learning, the source domain boasts abundant labeled training samples, whereas the target domain comprises only limited labeled training samples. Leveraging the information from the source domain can benefit the classification of the target domain. However, inconsistencies in land-cover classes between source and target domains may hinder knowledge transfer between domains. Fortunately, few-shot learning (FSL) provides an effective solution to this challenge. In recent years, meta-learning has gained widespread attention as a mainstream approach within FSL. This paper proposes a novel method for cross-domain heterogeneous HSI classification, called cross-domain meta-learning with feature alignment (CD-MFA). CD-MFA enhances the generalization performance of the inner-loop optimization by incorporating task-adaptive loss function. The adaptive weighting strategy is used in the outer-loop optimization to balance the classification losses of the source and target domains to learn more discriminative features. Additionally, by aligning the features of the source and target domains under the guidance of the Gaussian prior, the impact of domain shift can be mitigated. It is worth noting that CD-MFA is trained concurrently on both the source and target domains so that the two domains are will bound, thereby enhancing the effectiveness of knowledge transfer. Experimental results on four publicly available HSI datasets validate the effectiveness of CD-MFA.
Deep Inductive and Scalable Subspace Clustering via Nonlocal Contrastive Self-distillation
(Institute of Electrical and Electronics Engineers (IEEE), 2025-09-24) Zhu, W; Peng, B; Yan, WeiQi
Deep subspace clustering has demonstrated remarkable results by leveraging the nonlinear subspace assumption. However, it often encounters challenges in terms of computational cost and memory footprint in dealing with large-scale data due to its traditional single-batch training strategy. To address this issue, this paper proposes a deep subspace clustering framework that is regularized by nonlocal contrastive self-distillation, enabling a Deep Inductive and Scalable Subspace Clustering (DISSC) algorithm. In particular, our framework incorporates two subspace learning modules, namely subspace learning based on self-expression model and inductive subspace clustering. These modules generate affinities from different perspectives by extracting intermediate features from two augmentations of the input data using a weight-sharing neural network. By integrating the concept of self-distillation, our framework effectively exploits the clustering-friendly knowledge contained in these two affinities through a novel nonlocal contrastive prediction task, employing an empirical yet effective threshold. This allows the framework to facilitate complementary knowledge mining and scalability without compromising clustering performance. With an alternate branch that bypasses the self-expression computation, our framework can infer subspace membership of the out-of-sample data through the predicted soft labels, eliminating the need for ad-hoc postprocessing. In addition, the self-expression matrix computed using mini-batch data benefits from the distilled knowledge obtained from the inductive subspace clustering module, enabling our framework to scale to data of arbitrary size. Experiments conducted on large-scale MNIST, Fashion-MINST, STL-10, CIFAR-10 and Stanford Online Products datasets validate the superiority of the proposed DISSC algorithm over state-of-the-art subspace clustering methods.
A Novel Model-free Defense Scheme for Power Systems Stability Under Cyber Attacks
(Wiley, 2025-12-28) Oshnoei, S; Peykarporsan, R; Heidari, J; Mahboubi-Moghaddam, E; Lie, TT; Khooban, MH
The load frequency control (LFC) scheme, as a vital application in power systems' stability, makes the power system susceptible to cyber-attacks due to its dependence on information technologies and communication networks. This paper studies the LFC performance of Kundur's 4-unit-12-bus power system under false data injection (FDI) attacks. The available defence schemes are either based on the system's model or data-driven. The effectiveness of these schemes depends on the precise mathematical modelling or the extensive historical data of the power system. Therefore, it is necessary to design a defence strategy without depending on the mathematical model and the historical data of the system. To this end, this paper proposes a model-free resilient defence strategy, comprising a model-free detection scheme and an event-triggered mechanism. The presented detection scheme accomplishes the manipulated signal estimation using the measurement and control signals and compares the difference between the estimated and observed signals with a predefined threshold value. When the difference exceeds the threshold value, the detection scheme announces that an attack has occurred on the system. After detecting an attack, the event-triggered mechanism is activated to mitigate the attack's effect on the system frequency response. Accordingly, the event-triggered mechanism blocks the falsified signal and submits the estimated signal to the LFC controller. The presented scheme is independent of the system's mathematical model and historical data and can be employed in any cyber-physical power system. The design process of this strategy is simple and independent of the size and complexity of the power system. A deep reinforcement learning algorithm is also employed to tune the adjustable parameters of the proposed method. The real-time results obtained by the OPAL-RT simulator show that the developed scheme can timely identify FDI attacks and completely mitigate the attack's effect on the system's dynamic performance.
Finite Analysis of Self-Centring Beam-to-Column Subassemblies in Seismic-Resilient Steel MRF: Preliminary Evaluation of the Frame Expansion Effects
(ECCOMAS Proceedia, 2025-06-18) Roumieh, A; Kauntz Moderini, L; Elettore, E; Di Benedetto, S; Francavilla, AB; Latour, M; Gutierrez-Urzua, F; Pieroni, L; Ramhormozian, S; Simpson, B; Barbosa, AR; Rizzano, G; Grant, D; Ribeiro, F; Correia, AA; Freddi, F
Conventional code-based seismic design methods widely applied worldwide rely on the dissipation of seismic energy through construction damage. While this approach ensures life safety, it often results in significant post-earthquake damage, leading to substantial direct and indirect losses that affect communities’ resilience. To overcome these limitations, modern Earthquake Engineering is focusing on developing high-performance, cost-effective structures capable of withstanding design-level earthquakes with minimal socio-economic impact. In this context, the ERIES – SC-RESTEEL (Self-Centring seismic-RESilient sTEEL structures) project explores the structural response, repairability, resilience, and performance recovery of a steel self-centring Moment-Resisting Frame (MRF) incorporating friction devices and post-tensioned bars at column bases and beam-to-column joints. The project includes full-scale shaking table tests on a three-storey steel MRF at LNEC (Laboratório Nacional de Engenharia Civil) in Lisbon, Portugal. Moreover, the project investigates the response of the beam-to-column joint and the effect of the frame expansion due to the rocking of the beams through quasi-static cyclic tests on MRF subassemblies in Salerno, Italy. This paper illustrates the preparatory numerical work, including advanced Finite Element (FE) models in ABAQUS considering two configurations of the subassemblies, and investigates a solution to mitigate the frame expansion effects. The combined FE and experimental results provide crucial insights into the design of shaking table tests and the expected experimental outcomes.
Prompt-based Few-Shot Text Classification With Multi-Granularity Label Augmentation and Adaptive Verbalizer
(MDPI AG, 2026-01-08) Huang, Deling; Li, Zanxiong; Yu, Jian; Zhou, Yulong
Few-Shot Text Classification (FSTC) aims to classify text accurately into predefined categories using minimal training samples. Recently, prompt-tuning-based methods have achieved promising results by constructing verbalizers that map input data to the label space, thereby maximizing the utilization of pre-trained model features. However, existing verbalizer construction methods often rely on external knowledge bases, which require complex noise filtering and manual refinement, making the process time-consuming and labor-intensive, while approaches based on pre-trained language models (PLMs) frequently overlook inherent prediction biases. Furthermore, conventional data augmentation methods focus on modifying input instances while overlooking the integral role of label semantics in prompt tuning. This disconnection often leads to a trade-off where increased sample diversity comes at the cost of semantic consistency, resulting in marginal improvements. To address these limitations, this paper first proposes a novel Bayesian Mutual Information-based method that optimizes label mapping to retain general PLM features while reducing reliance on irrelevant or unfair attributes to mitigate latent biases. Based on this method, we propose two synergistic generators that synthesize semantically consistent samples by integrating label word information from the verbalizer to effectively enrich data distribution and alleviate sparsity. To guarantee the reliability of the augmented set, we propose a Low-Entropy Selector that serves as a semantic filter, retaining only high-confidence samples to safeguard the model against ambiguous supervision signals. Furthermore, we propose a Difficulty-Aware Adversarial Training framework that fosters generalized feature learning, enabling the model to withstand subtle input perturbations. Extensive experiments demonstrate that our approach outperforms state-of-the-art methods on most few-shot and full-data splits, with F1 score improvements of up to +2.8% on the standard AG’s News benchmark and +1.0% on the challenging DBPedia benchmark.
Bayesian Power Spectral Density Estimation for LISA Noise Based on Penalized Splines With a Parametric Boost
(American Physical Society (APS), 2026-01-09) Aimen, Nazeela; Maturana-Russel, Patricio; Vajpeyi, Avi; Christensen, Nelson; Meyer, Renate
Flexible and accurate noise characterization is crucial for the precise estimation of gravitational-wave parameters. We introduce a Bayesian method for estimating the power spectral density (PSD) of long, stationary time series, explicitly tailored for Laser Interferometer Space Antenna (LISA) data analysis. Our approach models the PSD as the geometric mean of a parametric and a nonparametric component, combining the knowledge from parametric models with the flexibility to capture deviations from theoretical expectations. The nonparametric component is expressed by a mixture of penalized B splines. Adaptive, data-driven knot placement, performed once at initialization, removes the need for a reversible-jump Markov chain Monte Carlo, while hierarchical roughness-penalty priors prevent overfitting. Validation on simulated autoregressive (AR) data of order 4 [AR(4)] demonstrates estimator consistency and shows that well-matched parametric components reduce the integrated absolute error compared to an uninformative baseline, requiring fewer spline knots to achieve comparable accuracy. Applied to one year of simulated LISA 𝑋-channel (univariate) noise, our method achieves relative integrated absolute errors of 𝒪⁡(10ˉ²), making it suitable for iterative analysis pipelines and multiyear mission data sets.

Browse

Recent Submissions