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 1830

TPR: Topology-Preserving Reservoirs for Generalized Zero-Shot Learning
(NeurIPS Proceedings, 2024-09-26) Chen, Hui; Liu, Yanbin; Ma, YongQiang; Zheng, Nanning; Yu, Xin
Pre-trained vision-language models (VLMs) such as CLIP have shown excellent performance for zero-shot classification. Based on CLIP, recent methods design various learnable prompts to evaluate the zero-shot generalization capability on a base-to-novel setting. This setting assumes test samples are already divided into either base or novel classes, limiting its application to realistic scenarios. In this paper, we focus on a more challenging and practical setting: generalized zero-shot learning (GZSL), i.e., testing with no information about the base/novel division. To address this challenging zero-shot problem, we introduce two unique designs that enable us to classify an image without the need of knowing whether it comes from seen or unseen classes. Firstly, most existing methods only adopt a single latent space to align visual and linguistic features, which has a limited ability to represent complex visual-linguistic patterns, especially for fine-grained tasks. Instead, we propose a dual-space feature alignment module that effectively augments the latent space with a novel attribute space induced by a well-devised attribute reservoir. In particular, the attribute reservoir consists of a static vocabulary and learnable tokens complementing each other for flexible control over feature granularity. Secondly, finetuning CLIP models (e.g., prompt learning) on seen base classes usually sacrifices the model's original generalization capability on unseen novel classes. To mitigate this issue, we present a new topology-preserving objective that can enforce feature topology structures of the combined base and novel classes to resemble the topology of CLIP. In this manner, our model will inherit the generalization ability of CLIP through maintaining the pairwise class angles in the attribute space. Extensive experiments on twelve object recognition datasets demonstrate that our model, termed Topology-Preserving Reservoir (TPR), outperforms strong baselines including both prompt learning and conventional generative-based zero-shot methods.
AI and Work Engagement: A Study of IT Professionals Through the Lens of Self-Determination Theory
(Springer Nature, 2025-09-01) Cajander, Åsa; Bergqvist, A; Clear, Tony; Daniels, Mats; Humble, N; Larusdottir, M; Normak, M; Oubhi, S
Artificial intelligence (AI) is reshaping work, presenting new challenges and opportunities for professionals across industries. This study, grounded in Self-Determination Theory (SDT), examines the impact of AI integration on work engagement among IT professionals. Through qualitative investigation, our research reveals that AI both augments and complicates professionals’ work lives, providing opportunities for growth while also demanding ongoing adaptation. Key findings indicate that AI tools like UiPath and GitHub Copilot enhance work efficiency by automating routine tasks, enabling professionals to concentrate on more complex aspects of their work and enhancing their perceived competence. However, this efficiency gain requires continuous learning and adaptation, posing challenges in maintaining engagement and mastery. These findings illuminate the complex balance between leveraging AI for increased efficiency and maintaining the intrinsic human elements of IT design, offering some insights for navigating AI integration in the workplace.
Knowledge Distillation With Differentiable Optimal Transport on Graph Neural Networks
(Springer Nature, 2025-07-22) Li, Mengyao; Liu, Yanbin; Chen, Ling
Knowledge distillation (KD) transfers knowledge from a large, well-trained teacher network to a smaller student network, improving student’s performance without extra computational costs. Traditional KD methods usually focus on the logits or intermediate features. However, they might overlook the inherent correlation and suffer from capacity gaps due to the distinct architectures of the student and teacher. Relation-based distillation methods try to bridge these correlation but usually require a large memory bank for loss computation, being less efficient. To overcome those limitations, we propose a novel and efficient Optimal Transport-based Graph Distillation (OTGD) method. First, OTGD constructs attributed graphs for the teacher and student respectively, which are then utilized to capture both individual and relational knowledge through graph neural networks (GNNs). Then, we devise an innovative differentiable optimal transport objective to distill the teacher knowledge before and after GNNs learning, effectively incorporating both the feature-level and correlation-level knowledge. Specifically, our optimal transport objective is solved by the Sinkhorn algorithm without relying on an extra memory bank. This design makes our method efficient and numerically stable. Comprehensive experiments conducted on two benchmark datasets with diverse network architectures, and demonstrate that OTGD outperforms the state-of-the-art methods.
Knowledge Transfer via Augmented Reality (AR): A Case Study of AR in Damage Control Equipment Maintenance Training in the Navy
(Springer, 2026-01-02) Strahan, Kelly; Konings, Daniel; Lovreglio, Ruggiero; Alam, Fakhrul
The art of transferring knowledge from one source to another is how humankind has built most of the vast wisdom we enjoy today. Knowledge sharing occurs at much higher rates today due to new digital ways of capturing and sharing knowledge. Recent literature suggests that extended reality-based training can benefit organizations. This paper explores the capabilities of Augmented Reality (AR), a subset of extended reality, in transferring technical knowledge within an organization. It is not well understood how the benefits of AR based training change depending on the complexity of the task. We attempt to address this gap by developing HoloLens2 based authored AR guides for four maintenance tasks of differing complexity and testing with twenty-three participants of differing levels of experience. Each participant, recruited from the Royal New Zealand Navy, performed two simple and two complex tasks. The participants used AR guides for one simple and one complex task. They completed the remaining two tasks with traditional paper instructions. The training effectiveness is assessed by time of completion of tasks, count of errors, and participants’ cognitive load. The participants using AR made fewer errors and reported a lower cognitive load for the training. However, the time spent on the AR training was longer for three of the four tasks.
Legibility vs. Extractability: Crafting Visual Defenses Against Automated OCR
(ACM Digital Library, 2026-01-02) Nguyen, Minh; T. P. Tran, Kien; Huynh, The Han
The rise of generative AI, particularly large language models (LLMs), has redefined problem-solving across domains—but it has also introduced new challenges in academic integrity. Students are increasingly using AI-powered Optical Character Recognition (OCR) tools to extract restricted content from screenshots, bypassing traditional safeguards that prevent copying and pasting. In this study, we investigate a set of visual defense techniques designed to counter automated OCR systems: (1) adversarial fonts crafted to disrupt character recognition, (2) color-based distortions that alter visual contrast, (3) animated interference lines that obstruct character boundaries, and (4) a novel cloud blur effect that dynamically follows the cursor to obscure localized text regions. We evaluate these strategies across multiple LLM-integrated OCR platforms—ChatGPT, DeepSeek, Claude, and Gemini. Our findings show that modern OCR tools remain largely unaffected by custom fonts, line obstructions, and color distortions. In contrast, the cloud blur technique significantly reduces OCR accuracy while preserving legibility for human readers. These results highlight dynamic, context-aware visual obfuscation as a promising and potentially future-proof solution for deterring AI-assisted text extraction. Cloud blur, in particular, emerges as the most effective approach, offering strong resistance to OCR while maintaining accessibility for legitimate human users. A live demo is available at https://cv.aut.ac.nz/nFonts.
Federated Learning and Data Mining-Based Botnet Attack Detection Framework for Internet of Things
(MDPI AG, 2026-03-02) Sudheera, Kalupahana Liyanage Kushan; Priyashan, Lokuge Lehele Gedara Madhuwantha; Pavithra, Oruthota Arachchige Sanduni; Aththanayake, Malwaththe Widanalage Tharindu; Sudasinghe, Piyumi Bhagya; Sankalpa, Wijethunga Gamage Chatum Aloj; Sandamali, Gammana Guruge Nadeesha; Chong, Peter Han Joo
Botnet attacks in Internet of Things (IoT) environments often occur as multi-stage campaigns, making early and reliable detection difficult across distributed and privacy-sensitive networks. Centralized detection approaches are often limited by heterogeneous traffic characteristics, severe data imbalance, and the need to aggregate large volumes of raw network data, raising scalability and privacy concerns. To address these challenges, this paper proposes FDA, a federated learning-based and data mining-driven framework for stage-aware botnet attack detection in IoT networks. FDA operates at network gateways, where anomalous traffic is first detected and then abstracted into compact and interpretable patterns using Frequent Itemset Mining (FIM). This pattern-based representation reduces noise and local traffic bias, enabling more robust learning across different IoT networks. Lightweight neural network models are trained locally at gateways, and a global model is learned through federated aggregation of model parameters, avoiding direct sharing of raw network data while enabling gateways to collaboratively learn evolving attack patterns across different IoT networks. Experimental results show that FDA achieves anomaly detection F1-scores above 99% across all gateways and multi-stage botnet attack classification F1-scores in the range of 48–49%, which are comparable to centralized machine-learning baselines while operating under decentralized and privacy-preserving constraints. Overall, FDA provides a practical, privacy-preserving, and effective solution for distributed botnet attack stage detection in real-world IoT deployments.
Leveraging AI and Technology for Holistic Asthma Management in the Pasifika Community
(Elsevier BV, 2026-02-24) Mirza, Farhaan; Jayamini, Widana Kankanamge Darsha; Lutui, Raymond; Pole, Kalesita; Matenga-Ikihele, Amio; Mu’aulama, Arieta Fa’apesolo; Chan, Amy Hai Yan
Asthma is a major health concern for Pasifika in New Zealand (NZ). This study explores the perspectives of Pasifika in NZ regarding the use of technology and Artificial Intelligence (AI) for asthma management. This study employed a qualitative research design, using purposive sampling to recruit eighteen Pasifika participants diagnosed with asthma. Data were analyzed using a general inductive approach. Six core themes were identified: (1) Awareness of AI; (2) Potential of AI; (3) Concerns of AI; (4) Digital divide in age groups; (5) General asthma management; and (6) Future design considerations. A holistic framework is postulated to enhance Pasifika asthma care using technology and AI. This study contributes to the sparse literature on Pasifika perspectives regarding AI-driven asthma management. Effective implementation requires ensuring accuracy, protecting privacy, improving access, and supporting families to confidently manage asthma. A balanced approach integrating technology with personal management is the key to effective asthma management.
OMNI-Dent: Towards an Accessible and Explainable AI Framework for Automated Dental Diagnosis
(Computer Vision Foundation, 2026-03-06) Jang, Leeje; Chiang, Yao-Yi; Hastings, Angela M; Pungchanchaikul, Patimaporn; Lucas, Martha B; Schultz, Emily C; Louie, Jeffrey P; Estai, Mohamed; Wang, Wen-Chen; Ip, Ryan HL; Huang, Boyen
Accurate dental diagnosis is essential for oral healthcare, yet many individuals lack access to timely professional evaluation. Existing AI-based methods primarily treat diagnosis as a visual pattern recognition task and do not reflect the structured clinical reasoning used by dental professionals. These approaches also require large amounts of expert-annotated data and often struggle to generalize across diverse real-world imaging conditions. In this paper, we present OMNI-Dent, a data-efficient and explainable diagnostic framework that incorporates clinical reasoning principles into a Vision-Language Model (VLM)-based pipeline. The framework operates on multi-view smartphone photographs, embeds diagnostic heuristics from dental experts, and guides a general-purpose VLM to perform tooth-level evaluation without dental-specific fine-tuning of the VLM. By leveraging the VLM's existing visual-linguistic capabilities, OMNI-Dent supports diagnostic assessment in settings where curated clinical imaging is unavailable. We design OMNI-Dent as an early-stage assistive tool to help users identify potential abnormalities and determine when professional evaluation may be needed, thereby offering a practical option for individuals with limited access to in-person care.
Design and Analysis of a 3D Frictional Mechanical Metamaterial for Efficient Energy Dissipation
(Wiley, 2024-10-30) Jeong, E; Calius, E; Ramezani, M
This study introduces a novel frictional mechanical metamaterial composed of a central hexagon or re-entrant honeycomb frame, a lower section with four tapered columns, and an upper portion with a blade shape. When subjected to an external uniaxial force, the 3D structure of the metamaterial utilizes sliding interactions to dissipate frictional energy. The mechanical properties of the proposed metamaterial, such as load-displacement relationships, hysteresis area, and peak force, can be fine-tuned by adjusting geometric parameters and constituent materials. Extensive analysis is conducted through experimental compression tests, finite element (FE) simulations, and theoretical modeling. Comparative assessments of the metamaterial's energy dissipation performance demonstrated a good agreement between experimental and simulation results, with minor variations observed for deeper compression cycles. The proposed metamaterial offers the potential for superior elastic energy absorption and dissipation, making it a promising solution for applications requiring repeated energy dissipation or damping under cyclical loads while maintaining a lightweight profile.
Sintering Parameter Investigation for Bimetallic Stainless Steel 316L/ Inconel 718 Composite Printed by Dual-nozzle Fused Deposition Modeling
(Emerald, 2024-07-25) Jiang, CP; Masrurotin, M; Ramezani, M; Wibisono, AT; Toyserkani, E; Macek, W
Purpose: Fused deposition modeling (FDM) nowadays offers promising future applications for fabricating not only thermoplastic-based polymers but also composite PLA/Metal alloy materials, this capability bridges the need for metallic components in complex manufacturing processes. The research is to explore the manufacturability of multi-metal parts by printing green bodies of PLA/multi-metal objects, carrying these objects to the debinding process and varying the sintering parameters. Design/methodology/approach: Three different sample types of SS316L part, Inconel 718 part and bimetallic composite of SS316L/IN718 were effectively printed. After the debinding process, the printed parts (green bodies), were isothermally sintered in non-vacuum chamber to investigate the fusion behavior at four different temperatures in the range of 1270 °C−1530 °C for 12 h and slowly cooled in the furnace. All samples was assessed including geometrical assessment to measure the shrinkage, characterization (XRD) to identify the crystallinity of the compound and microstructural evolution (Optical microscopy and SEM) to explore the porosity and morphology on the surface. The hardness of each sample types was measured and compared. The sintering parameter was optimized according to the microstructural evaluation on the interface of SS316L/IN718 composite. Findings: The investigation indicated that the de-binding of all the samples was effectively succeeded through less weight until 16% when the PLA of green bodies was successfully evaporated. The morphology result shows evidence of an effective sintering process to have the grain boundaries in all samples, while multi-metal parts clearly displayed the interface. Furthermore, the result of XRD shows the tendency of lower crystallinity in SS316L parts, whilst IN718 has a high crystallinity. The optimal sintering temperature for SS316L/IN718 parts is 1500 °C. The hardness test concludes that the higher sintering temperature gives a higher hardness result. Originality/value: This study highlights the successful sintering of a bimetallic stainless steel 316 L/Inconel 718 composite, fabricated via dual-nozzle fused deposition modeling, in a non-vacuum environment at 1500 °C. The resulting material displayed maximum hardness values of 872 HV for SS316L and 755.5 HV for IN718, with both materials exhibiting excellent fusion without any cracks.
Fabrication of Translucent Graded Dental Crown Using Zirconia-Yttrium Multi-Slurry Tape Casting 3D Printer
(Elsevier BV, 2024-01-26) Romario, YS; Bhat, C; Ramezani, M; Pasang, T; Chen, Z; Jiang, CP
This paper aims to fabricate functionally graded dental crown using a multi-slurry tape casting additive manufacturing technology. The different luminescence of the dental crown was obtained with different composition of zirconia and yttria. Zirconia with tunable mechanical properties and translucency are obtained by adding 3, 3.5, 4, 4.5, and 5 mol% of yttrium oxide to zirconia powder. After obtaining the printable slurry with maximum solid loading, the green bodies are prepared using the in-house built high-speed multi-ceramic tape casting technology. They are later sintered with two-stage sintering method. After the successful fabrication, the mechanical properties and translucency of the specimens were evaluated with Vickers hardness, three-point bending and translucency parameter tests. Finally, an FGM tooth crown with five photocurable slurries is proposed to demonstrate the translucent gradient effect of sintered part. The solid loading of 80% zirconia and 20% resin delivered samples without any surface cracks. The shrinkage ratio analysis showed that the sintered sample dimension was reduced by 20%, 20%, and 23% along X, Y, and Z directions. The samples fabricated with 3% yttrium oxide to zirconia delivered excellent hardness (1687 HV) and flexural strength (650.6 MPa). However, the relative luminescence increased with increasing the yttrium oxide for 3–5 mol%. With the optimized process parameters, the proposed dental crown is fabricated and analyzed for their shrinkage ratio, mechanical, and translucency properties. The study proposes the potential of fabricating customized dental crown with gradient translucent appearance.
POI Recommendation for Random Groups Based on Cooperative Graph Neural Networks
(Elsevier BV, 2024-02-05) Liu, Zhizhong; Meng, Lingqiang; Sheng, Quan Z; Chu, Dianhui; Yu, Jian; Song, Xiaoyu
Group Point-of-Interests (POI) recommendation devotes to find the optimal POIs for groups, which has extracted extensive attention. This work first brings forward a novel POI recommendation model for random groups based on Cooperative Graph Neural Networks (named as CGNN-PRRG). We have done three innovative work. (1) We propose a new fitted presentation learning method for generating the fitted representations of random groups. (2) To conquer the cold start issues in recommending POI for a new random group, we propose to take similar users’ (which have the similar representations with that of the random group) POI interaction data as the learning data. (3) We propose an Edge-learning enhanced Bipartite Graph Neural Network (EBGNN) to learn similar users’ POI comprehensive interaction preferences. Specially, EBGNN can learn the information on the edges of the graph. Meanwhile, we propose to learn similar users’ POI transfer preferences with the Session-based Graph Neural Networks (SRGNN). We verify our proposed model on the three public benchmark datasets (Foursquare, Gowalla and Yelp), which contain 124,933 to 860,888 POI check-in records. The comparison between our proposed model and ten representative baseline models demonstrates the outstanding performance of CGNN-PRRG. In terms of Precision@K and NDCG@K, our model achieves about 24.9% and 62.5% improvement compared with the best baseline models on the three benchmark datasets averagely. Adequate ablation experiments prove the effectiveness of the fitted representation generation method, similar users’ POI comprehensive interaction preferences learning method and the method for overcoming the cold start problem. The source code of the CGNN-PRRG model is available on github1.
Autonomous Fingerprinting and Large Experimental Data Set for Visible Light Positioning
(MDPI AG, 2021-05-08) Glass, T; Alam, F; Legg, M; Noble, F
This paper presents an autonomous method of collecting data for Visible Light Positioning (VLP) and a comprehensive investigation of VLP using a large set of experimental data. Received Signal Strength (RSS) data are efficiently collected using a novel method that utilizes consumer grade Virtual Reality (VR) tracking for accurate ground truth recording. An investigation into the accuracy of the ground truth system showed median and 90th percentile errors of 4.24 and 7.35 mm, respectively. Co-locating a VR tracker with a photodiode-equipped VLP receiver on a mobile robotic platform allows fingerprinting on a scale and accuracy that has not been possible with traditional manual collection methods. RSS data at 7344 locations within a 6.3 × 6.9 m test space fitted with 11 VLP luminaires is collected and has been made available for researchers. The quality and the volume of the data allow for a robust study of Machine Learning (ML)-and channel model-based positioning utilizing visible light. Among the ML-based techniques, ridge regression is found to be the most accurate, outperforming Weighted k Nearest Neighbor, Multilayer Perceptron, and random forest, among others. Model-based positioning is more accurate than ML techniques when a small data set is available for calibration and training. However, if a large data set is available for training, ML-based positioning outperforms its model-based counterparts in terms of localization accuracy.
Enhanced Particle Swarm Optimisation Algorithms for Multiple-input Multiple-output System Modelling Using Convolved Gaussian Process Models
(Inderscience Publishers, 2018-08-02) Cao, G; Lai, EMK; Alam, F
Convolved Gaussian process (CGP) can capture the input-output correlation, and the correlation of multiple outputs. This is beneficial to the modelling problem of multiple-input multiple-output (MIMO) systems. One key issue of CGP is the learning of hyperparameters from input-output observations. This is typically performed by maximising the log-likelihood (LL) function using gradient based approaches. However, the LL value is not a reliable indicator for judging the quality of intermediate models. We address this issue by minimising the model output error instead. In addition, three enhanced particle swarm optimisation (PSO) algorithms are proposed to solve the optimisation problem because gradient based approaches often get stuck in local optima. The simulation results on numerical linear and nonlinear systems demonstrate the effectiveness of minimising the model output error to learn hyperparameters, and the better performance of using enhanced PSOs compared to gradient based approaches.
LSD Microdosing for Major Depressive Disorder: Mood and Pharmacokinetic Outcomes From a Phase 2a Trial
(Elsevier, 2026-02-17) Daldegan-Bueno, Dimitri; Donegan, Carina Joy; Sumner, Rachael; Forsyth, Anna; Jeong, Soo Hee; Evans, William; Alshakhouri, Malak; Murphy, Robin J; Reynolds, Lisa; Hoeh, Nicholas; Allen, Nathan; Sundram, Frederick; Menkes, David; Muthukumaraswamy, Suresh
INTRODUCTION: Despite growing interest in microdosed psychedelics, clinical trial evidence remains limited. We present daily mood, subjective perception of effects, and pharmacokinetics from an 8-week regimen of microdosed lysergic acid diethylamide (LSD) as a treatment for major depressive disorder in an open-label trial in which participants reported a mean symptom reduction of 60%. METHODS: Participants took 16 sublingual LSD doses: 8 μg onsite, with bloods collected at eight time-points, then twice weekly at home with titration (6-20 μg). Pharmacokinetic parameters were estimated using non-compartmental and compartmental modelling. Daily questionnaires were used to assess depression severity with the self-reported Hamilton Depression Rating Scale (HAMD6), and mood with visual analogue scales (VAS). Drug effects were recorded with VAS scales on each dosing day. Linear mixed models were used to compare dosing days to one- and two-day post-dosing, and to identify linear trends (tolerance/sensitisation) of drug effects. RESULTS: Nineteen participants (males n = 15, 79%) received the intervention. Daily VAS indicated increased scores of mood-related states (e.g., more creative, happier) on dosing days (p = 0.009 to 0.039), but not in depression (p = 0.291). There was no indication of tolerance or sensitisation (p > 0.081). Non-compartmental AUC0-tlast was 836 ± 319 pg.h/mL, Cmax 212 ± 77.7 pg/mL and Tmax 1.17 ± 0.56 h. DISCUSSION: Results suggest short-term improvements in mood following microdosed LSD in people with depression, warranting confirmation in controlled trials. It provides the pharmacokinetic parameters of 8 μg of LSD in a sample of people with depression and indicates no tolerance or sensitisation to repeated microdoses of LSD, despite incremental dose titration.
Full-Scale Simulated Seismic Field-Testing of a Post-Tensioned Glue Laminated Timber Portal Frame Structure Deploying Traditional Māori Construction Techniques
(Taylor and Francis Group, 2026-02-24) Vercoe, Sonny; Hōete, Anthony; Ingham, Jason; Beskhyroun, Sherif
Testing was undertaken to investigate the viability of a traditional Māori post-tensioned construction technique that incorporated modern glue-lamination manufacturing to withstand seismic loadings. A feature of the form-fit connections was that the timber members were seated into pockets. A companion numerical model was developed to confirm aspects not readily measured in the field. Vertical loading was applied at the apex via water-filled containers. Seismic loading was simulated by applying a horizontal loading system at the apex via a winch and quick-release system, involving seven pseudo-static semi-cyclic tests followed by three snap-back tests. It was established that member demands remained within capacity, that apex vertical and horizontal displacements remained within code-defined deflection limits, and that the measured damping was 7.3%. Findings are also reported for mode shapes and joint rotational stiffness. Novel aspects of the reported research included full-scale field testing at a remote site, collaborations with the local indigenous Māori community, the incorporation of both traditional and modern post-tensioned timber construction techniques, and the intentional use of a methodology for simulated loading which could be undertaken within a community setting.
SIFT-SNN for Traffic-Flow Infrastructure Safety: A Real-Time Context-Aware Anomaly Detection Framework
(MDPI AG, 2026-01-31) Rathee, Munish; Bačić, Boris; Doborjeh, Maryam
Automated anomaly detection in transportation infrastructure is essential for enhancing safety and reducing the operational costs associated with manual inspection protocols. This study presents an improved neuromorphic vision system, which extends the prior SIFT-SNN (scale-invariant feature transform–spiking neural network) proof-of-concept by incorporating temporal feature aggregation for context-aware and sequence-stable detection. Analysis of classical stitching-based pipelines exposed sensitivity to motion and lighting variations, motivating the proposed temporally smoothed neuromorphic design. SIFT keypoints are encoded into latency-based spike trains and classified using a leaky integrate-and-fire (LIF) spiking neural network implemented in PyTorch. Evaluated across three hardware configurations—an NVIDIA RTX 4060 GPU, an Intel i7 CPU, and a simulated Jetson Nano—the system achieved 92.3% accuracy and a macro F1 score of 91.0% under five-fold cross-validation. Inference latencies were measured at 9.5 ms, 26.1 ms, and ~48.3 ms per frame, respectively. Memory footprints were under 290 MB, and power consumption was estimated to be between 5 and 65 W. The classifier distinguishes between safe, partially dislodged, and fully dislodged barrier pins, which are critical failure modes for the Auckland Harbour Bridge’s Movable Concrete Barrier (MCB) system. Temporal smoothing further improves recall for ambiguous cases. By achieving a compact model size (2.9 MB), low-latency inference, and minimal power demands, the proposed framework offers a deployable, interpretable, and energy-efficient alternative to conventional CNN-based inspection tools. Future work will focus on exploring the generalisability and transferability of the work presented, additional input sources, and human–computer interaction paradigms for various deployment infrastructures and advancements.
Standalone DC Microgrids: Planning, Operation and Uncertainty Management
(Elsevier BV, 2026-01-20) Jayasinghe, H; Gunawardane, K; Hossain, MA; Zamora, R
Standalone power systems in remote areas have traditionally relied on continuously operating fossil fuel generators, leading to high operational costs, reduced efficiency, and substantial carbon emissions. Standalone direct current (DC) microgrids have emerged as a promising alternative due to their lower conversion losses, improved integration of renewable energy sources (RES), and enhanced compatibility with modern DC-native loads and storage technologies. Despite these advantages, the planning, operation, and uncertainty management of standalone DC microgrids remain technically challenging. Intermittent RES generation, stochastic load behaviour, lack of mature standards, and complex control requirements introduce significant design and operational challenges. While numerous studies have proposed methods to address issues in sizing, optimisation, control, energy management, and uncertainty management, a comprehensive and structured review that connects these aspects across the full lifecycle of DC microgrid development is still lacking. This article addresses this gap by providing a systematic review of the state-of-the-art in planning methodologies, operational strategies, and uncertainty management techniques for standalone DC microgrids. The review synthesises theoretical frameworks and practical implementations, critically evaluates existing approaches by identifying their strengths and limitations, and highlights the interdependencies among planning, real-time operation, and uncertainty mitigation. Finally, the article outlines key research challenges and future opportunities to support the reliable, cost-effective, and sustainable deployment of standalone DC microgrids. The novelty of this study lies in its integrated perspective spanning planning, operational control, and uncertainty management, offering valuable guidance for researchers, system designers, and practitioners.
Optimizing the Resistivity of Colloidal SnO₂ Thin Films by Ion Implantation and Annealing
(Elsevier BV, 2024-10-28) Yusuf, AS; Markwitz, M; Chen, Z; Ramezani, M; Kennedy, JV; Fiedler, H
Tin oxide (SnO₂) is a critical material for a wide range of applications, such as in perovskite solar cells, gas sensors, as well as for photocatalysis. For these applications the transparency to visible light, high availability, cheap fabrication process and high conductivity of SnO₂ benefits its commercial deployment. In this paper, we demonstrate that the resistivity of widely colloidal SnO₂ can be reduced by noble gas ion beam modification. After low energy argon implantation with a fluence of 4×10¹⁵ at.cm⁻² at 25keV and annealing at 200°C in air, the resistivity of as-deposited film was reduced from (178±6)μΩcm to (133±5)μΩcm, a reduction of 25%. Hall effect measurements showed that the primary cause of this is the increase in carrier concentration from (8.1±0.3)×10²⁰ cm⁻³ to (9.9±0.3)×10²⁰ cm⁻³. Annealing at 200°C resulted in the removal of defect clusters introduced by implantation, while annealing at 300°C resulted in the oxidation of the films, increasing their resistivity. The concentration of oxygen vacancy defects can be controlled by a combination of low energy noble gas ion implantation and annealing, providing promising performance increases for potential applications of SnO₂ where a low resistivity is crucial.
Lateralized Learning for Multi-class Visual Classification Tasks
(Institute of Electrical and Electronics Engineers (IEEE), 2026-02-03) Siddique, Abubakar; Browne, Will N; Grimshaw, Gina M
The majority of computer vision algorithms fail to find higher-order (abstract) patterns in an image so they are not robust against adversarial attacks. Deep learning considers each input pixel in a homogeneous manner such that different parts of a locality-sensitive hashing table are often not connected, meaning higher-order patterns are not discovered. Hence, these systems are sensitive to noisy, irrelevant, and redundant data, leading to wrong predictions with high confidence. Adversarial attacks exploit this vulnerability by generating deceptive inputs that mislead AI systems. In contrast, human vision is rarely susceptible to adversarial attacks. Vertebrate brains afford heterogeneous knowledge representation through lateralization, enabling modular learning at different levels of abstraction. This work aims to verify the effectiveness, scalability, and robustness of a lateralized approach to real-world problems that contain noisy, irrelevant, and redundant data. Two well-known and widely used adversarial attacks, the Fast Gradient Sign Method and the Iterative Adversarial Technique, are applied to generate corrupted test images. The experimental results on multi-class (200 classes) image classification tasks demonstrate that the proposed system effectively captures hierarchical knowledge representations, enhancing its robustness. Crucially, the lateralized system outperformed four state-of-the-art deep learning-based systems for the classification of normal and adversarial images by 19.05% − 41.02% and 1.36% − 49.22%, respectively.

Browse

Recent Submissions