Conference outputs
Permanent URI for this collection
Browse
Recent Submissions
Conference output Failure Mechanisms of Multiple External Column Removals in RC Flat Plate StructuresZhao, ZQ; Guan, H; Xue, HZ; Li, Y; Gilbert, BP; Chouw, Nawawi; Zhang, Chunwei (26th Australasian Conference on the Mechanics of Structures and Materials, 2024)In a reinforced concrete (RC) flat slab structure, the external columns including corner and edge columns on the building perimeter are likely to be more susceptible to abnormal and accidental loads (such as vehicle impact, explosion, fire, etc.) than the interior ones, causing localised damage and a potential progressive collapse of the overall structure. Current research has mostly focused on the case of a single column removal, but due to the uncertainty of the accidental events, the likelihood of multiple columns being simultaneously destroyed is also high. This study will develop a 3D high-fidelity finite element model using the LS-DYNA program, to investigate and analyse the failure behaviour of three 2 × 2 bay RC flat plate substructures with single, two, and multiple column removals: (i) removal of a corner column only (Model C), the accuracy of which will be validated with our previous test results, (ii) removal of a corner column and an adjacent edge column (Model C_1e), and (iii) removal of a corner column and two adjacent edge columns (Model C_2e). The flexural failure, punching shear failure, and post-punching mechanisms of the models will be discussed based on their crack patterns, deformed shapes, and load-deflection responses.
Conference output Evaluating GPT's Programming Capability Through CodeWars' KatasZhang, Zizhuo; Wen, Lian; Zhang, Shaoyang; Chen, David; Jiang, Yanfei; Cao, C; Chen, H; Zhao, L; Arshad, J; Asyhari, T; Wang, Y (17th International Conference on Knowledge Science, Engineering and Management (KSEM), 2024)Understanding the capabilities and limitations of programming-oriented AI models is crucial. This paper evaluates the programming proficiency of GPT-3.5 and GPT-4 using Codewars coding problems of varying difficulty. The experiments reveal a distinct boundary at the 3kyu level, beyond which these models struggle. This led to proposing a complexity measure that includes problem difficulty and solution time. The research emphasizes the need for validation and creative thinking in AI models to better emulate human problem-solving. Future work aims to refine the complexity measure, enhance AI capabilities, and develop an objective programming problem difficulty measure. These insights are valuable for advancing AI programming and problem-solving abilities.
Conference output Applying Continuous Formal Methods to Cardano (Experience Report)Chapman, James; Bailly, Arnaud; Vinogradova, Polina; Sperber, M; Stevens, P (FUNARCH '24: 2nd ACM SIGPLAN International Workshop on Functional Software Architecture, 2024)Cardano is a Proof-of-Stake cryptocurrency with a market capitalisation in the tens of billions of USD and a daily volume of hundreds of millions of USD. In this paper we reflect on applying formal methods, functional architecture and Haskell to building Cardano. We describe our strategy, projects, lessons learned, the challenges we face, and how we propose to meet them.
Conference output Meshfree Model for Porous Seabed Response Under Combined Earthquake and Random Wave LoadingHan, Shuang; Zhang, Jisheng; Tsai, ChiaCheng; Jeng, Dong-Sheng (ASME 43rd International Conference on Ocean, Offshore and Arctic Engineering (OMAE), 2024)In the context of offshore structure design, the consideration of earthquake and wave loadings is paramount due to their pivotal role as natural dynamic forces. These dynamic forces can induce vibrations in the pore water pressure, thereby precipitating seabed instability. However, the investigations of the earthquake-induced seabed behaviour are limited. Furthermore, contemporary studies on seismic seabed response often neglect the concurrent impact of ocean waves. Previous research predominantly relies on conventional mesh-based methods, such as the finite element method. To address the limitations inherent in such methods, such as computational time and mathematical complexity, this study employs a meshfree method based on the “u – p” approximation. The research assesses soil response under the Japan 311 earthquake and random wave loading in both time and frequency domains. Numerical findings indicate that earthquake-induced acceleration is notably amplified by the seabed foundation, particularly in the horizontal direction. The presence of wave loading significantly alters the development of pore pressure, yet it exerts no discernible impact on earthquake-induced acceleration in the seabed. Consequently, the contribution of random waves to seismic-induced seabed response analysis cannot be disregarded.
Conference output Mitochondrial translation is the primary determinant of secondary mitochondrial complex I deficienciesCunatova, Kristyna; Vrbacky, Marek; Puertas-Frias, Guillermo; Alan, Lukas; Vanisova, Marie; Saucedo-Rodriguez, Maria Jose; Houstek, Josef; Fernandez-Vizarra, Erika; Neuzil, Jiri; Pecinova, Alena; Pecina, Petr; Mracek, Tomas (Biochimica et Biophysica Acta (BBA) - Bioenergetics, EBEC2024: 22nd European Bioenergetics Conference, 2024)Individual complexes of the mitochondrial oxidative phosphorylation system (OXPHOS) are not linked solely by their function; they also share dependencies at the maintenance/assembly level, where one complex depends on the presence of a different individual complex. Despite the relevance of this ‘interdependence’ behavior for mitochondrial diseases, its true nature remains elusive. To understand the mechanism that can explain this phenomenon, we examined the consequences of the aberration of different OXPHOS complexes in human cells. We demonstrate here that complete disruption of each of the OXPHOS complexes resulted in a perturbation in energy deficiency sensing pathways, including the integrated stress response (ISR) pathway. The secondary decrease of complex I (CI) level was triggered by both complex IV and complex V deficiency, and it was independent of ISR signaling. On the other hand, we identified the unifying mechanism behind CI downregulation in the downregulation of mitochondrial ribosomal proteins and, thus, mitochondrial translation. We conclude that the secondary CI defect is due to mitochondrial protein synthesis attenuation, while the responsible signaling pathways could differ based on the origin of the OXPHOS defect.
Funded by the Czech Science Foundation (22-21082S, 21-18993S), Czech Health Research Council (NU22-01-00499), and Next Generation EU project National Institute for Research of Metabolic and Cardiovascular Diseases (Programme EXCELES, ID LX22NPO5104).
Conference output Generation of airborne particles toward inhalation drug delivery via electro-neutralization electrosprayVu, HD; Nguyen, TD; Vu, TH; Mai, LN; Van Anh Hoang, T; Tran, DDH; Nguyen, TH; Zhu, Y; Dao, DV; Thanh Dau, V (2024 IEEE 19th International Conference on Nano/Micro Engineered and Molecular Systems (NEMS), 2024)This paper reports a one-step aerosol generation and delivery method based on electrohydrodynamic atomization (EHDA) for pulmonary drug delivery. By using inflight electro-neutralization electrospray (IFENE), the surface charge of particles was significantly reduced, allowing the generated aerosols to travel freely in open space while still having the desirable size. The IFENE method delivery ability was demonstrated by generating particles from different types of polymeric solution. The results showed that particles from IFENE method had the size distribution lying in the respirable range (lower than 5 μm) and surface charge significantly lower than normal electrospray and other inhalation devices. Along with the simple design, IFENE shows great potential to become a one-step pulmonary drug delivery device for various biomedical applications.
Conference output Experimental Analysis of Hybrid Timber-Steel Connections Under Quasi-Static LoadingThevarajah, BE; Remennikov, AM; Ngo, TD; Guan, H; Gilbert, BP; Chouw, Nawawi; Zhang, Chunwei (26th Australasian Conference on the Mechanics of Structures and Materials, 2024)Engineered Wood Products (EWP) are regaining popularity in the construction industry as a solution to the high carbon footprint of conventional construction materials, due to their low carbon footprint and high potential for recycling at the end-of-life cycle. Due to the brittle behaviour and low stiffness of the EWP structural members and low ductility in their connections, an accidental or deliberate extreme load to a critical structural member could initiate a localised failure which could lead to building collapse. This paper aims to investigate the behaviour of a hybrid timber-steel connection as a key to improving ductility. An experimental study was conducted on six specimens involving two types of connections subjected to quasi-static loading conditions. The first connection type is two steel fin plates placed in the middle of the LVL beam and the second is two steel fin plates placed on each side of the LVL beam and bolted together. The effects of the thickness of the steel plates on the behaviour were examined with 3 varying thick fin plates for each connection type. A 3D Digital Image Correlation (DIC) system was employed to obtain a three-dimensional strain field and review damage modes. The effects of the thickness of the fin plates and the fin plate arrangement on the ductile behaviour of the hybrid timber-steel connection are discussed in detail.
Conference output Bidfuse: Harnessing Bi-Directional Attention with Modality-Specific Encoders for Infrared-Visible Image FusionXing, Wangzhi; Chen, Diqi; Islam, Mohammad Aminul; Zhou, Jun (2024 IEEE International Conference on Image Processing (ICIP), 2024)Infrared-visible image fusion aims to utilize the distinct advantages of each modality to provide a more comprehensive representation than either one could offer. Current state-of-the-art methods segregate encoded features into modality-specific features and modality-independent features. However, this segregation often fails to effectively isolate feature representation from different modalities, which causes possible information loss, rendering overly complex and ultimately unnecessary design. To tackle this issue, we propose BIDFuse, a novel two-stage bi-directional fusion network, designed to leverage the unique features from both modalities without explicit feature separation. We first use two encoders to specifically extract critical information from the two input images. Additionally, a cross-feeding mechanism is implemented to feed the features from one encoder directly into the input stream of the other decoder, enhancing the reconstruction process with information from both sources. Then, the bi-directional attention module is designed to fuse the features from both modalities and generate a fused image. Competitive experimental results demonstrate the effectiveness of our method for image fusion on the MSRS dataset and for low-light object detection on the M3FD dataset.
Conference output Ambiguities in Defeasible Logic: A Computational Efficient Framework and AlgorithmGovernatori, G; Olivieri, F; Kirrane, Sabrina; Šimkus, Mantas; Soylu, Ahmet; Roman, Dumitru (8th International Joint Conference, RuleML+RR 2024, 2024)We present a Defeasible Logic variant able to incorporate two different and conflicting facets of non-monotonic reasoning: ambiguity blocking and ambiguity propagation. The resulting logic is conservative about the two notions. We investigate the logical properties and we present efficient algorithms for the logic.
Conference output Stress concentration factors (SCFs) for multi-planar tubular KK-joints of jacket substructures in offshore wind turbines (OWTs)Ahmadi, H; Karampour, H; Atalo, AA (34th International Ocean and Polar Engineering Conference, 2024)Although the investigation on the effect of loaded out-of-plane braces on the values of the stress concentration factor (SCF) in offshore tubular joints has been the objective of numerous research works, a number of quite important cases still exist that have not been studied thoroughly due to the diversity of joint types and loading conditions. One of these cases is the multi-planar tubular KK-joint subjected to axial loading. Tubular KK-joints are among the most common joint types in jacket substructure of offshore wind turbines (OWTs). In the present research, data extracted from the stress analysis of 243 finite element (FE) models, verified against available experimental data, was used to study the effects of geometrical parameters on the chord-side SCFs in multi-planar tubular KK-joints subjected to axial loading. Parametric FE study was followed by a set of nonlinear regression analyses to develop three new SCF parametric equations for the fatigue analysis and design of axially-loaded multi-planar KK-joints.
Conference output Homebirth is the Ultimate Scope of Practice. Why homebirth midwives continue to practice without professional indemnity insuranceMitchell, Jennifer; Gabriel, Laura; Hastie, Carolyn; Donnellan-Fernandez, Roslyn (Women and Birth, 2024 Australian College of Midwives National Conference: Moving Midwifery Forward, 2024)Background: Professional indemnity insurance for intrapartum care has not been available to Australian homebirth midwives since the collapse of Health Insurance Holdings in 2001. This puts midwives providing homebirth at significant professional and personal financial risk. The findings from this study will increase our knowledge of how Australian homebirth midwives operate their practice from a risk perspective, and why they continue to provide homebirth services without professional indemnity insurance.
Method: This research employed a phenomenological approach using semi-structured interviews to obtain in-depth insight into the views and experiences of Australian homebirth midwives.
Findings: Preliminary analysis of interview transcripts indicates that Midwives in private practice offer homebirth because it is the ultimate in working to scope of practice. Midwives report a deep connection to their community and value their relationships with women and families. The presentation will outline novel and unique insights into Midwives’ perspectives on risk, insurance and working to their full scope of practice. While some Midwives in this study reporting wanting insurance to protect their assets, others expressed concern about insurance restricting their practice.
How homebirth Midwives assess for and manage risk in their practice will be discussed, as well as their experiences supporting women choosing care outside of guidelines. The challenges and benefits of having admitting rights to local hospitals will also be explored. While some midwives described positive relationships with collaborating professionals, others reported being accused of ‘enabling’ women.
Conclusion: This Australian study is the first to identify and describe the perspectives of Australian homebirth midwives on the lack of intrapartum professional indemnity insurance. It is hoped that future research can further explore the issues identified so that the voices of homebirth midwives are heard.
Conference output How Much Time Does a Photon Spend as an Atomic Excitation Before Being Transmitted?Thompson, K; Li, K; Angulo, D; Nixon, VM; Sinclair, J; Sivakumar, AV; Wiseman, HM; Steinberg, AM (CLEO 2024: Conference on Lasers and Electro-Optics, 2024)We show that if a photon is transmitted through an atom cloud, the time it spent as an atomic excitation along the way—as measured by weakly probing the atoms— is equal to the group delay, which can be negative.
Conference output DarkFed: A Data-Free Backdoor Attack in Federated LearningLi, M; Wan, W; Ning, Y; Hu, S; Xue, L; Zhang, LY; Wang, Y; Larson, Kate (IJCAI-24: Thirty-Third International Joint Conference on Artificial Intelligence, 2024)Federated learning (FL) has been demonstrated to be susceptible to backdoor attacks. However, existing academic studies on FL backdoor attacks rely on a high proportion of real clients with main task-related data, which is impractical. In the context of real-world industrial scenarios, even the simplest defense suffices to defend against the state-of-the-art attack, 3DFed. A practical FL backdoor attack remains in a nascent stage of development. To bridge this gap, we present DarkFed. Initially, we emulate a series of fake clients, thereby achieving the attacker proportion typical of academic research scenarios. Given that these emulated fake clients lack genuine training data, we further propose a data-free approach to backdoor FL. Specifically, we delve into the feasibility of injecting a backdoor using a shadow dataset. Our exploration reveals that impressive attack performance can be achieved, even when there is a substantial gap between the shadow dataset and the main task dataset. This holds true even when employing synthetic data devoid of any semantic information as the shadow dataset. Subsequently, we strategically construct a series of covert backdoor updates in an optimized manner, mimicking the properties of benign updates, to evade detection by defenses. A substantial body of empirical evidence validates the tangible effectiveness of DarkFed.
Conference output Graph Attention Network with High-Order Neighbor Information Propagation for Social RecommendationXiong, F; Sun, H; Luo, G; Pan, S; Qiu, M; Wang, L; Larson, Kate (IJCAI-24: Thirty-Third International Joint Conference on Artificial Intelligence, 2024)In recommender systems, graph neural networks (GNN) can integrate interactions between users and items with their attributes, which makes GNN-based methods more powerful. However, directly stacking multiple layers in a graph neural network can easily lead to over-smoothing, hence recommendation systems based on graph neural networks typically underutilize higher-order neighborhoods in their learning. Although some heterogeneous graph random walk methods based on meta-paths can achieve higher-order aggregation, the focus is predominantly on the nodes at the ends of the paths. Moreover, these methods require manually defined meta-paths, which limits the model’s expressiveness and flexibility. Furthermore, path encoding in graph neural networks usually focuses only on the sequence leading to the target node. However, real-world interactions often do not follow this strict sequence, limiting the predictive performance of sequence-based network models. These problems prevent GNN-based methods from being fully effective. We propose a Graph Attention network with Information Propagation path aggregation for Social Recommendation (GAIPSRec). Firstly, we propose a universal heterogeneous graph sampling framework that does not require manually defining meta-paths for path sampling, thereby offering greater flexibility. Moreover, our method takes into account all nodes on the aggregation path and is capable of learning information from higher-order neighbors without succumbing to over-smoothing. Finally, our method utilizes a gate mechanism to fuse sequential and non-sequential dependence in encoding path instances, allowing a more holistic view of the data. Extensive experiments on real-world datasets show that our proposed GAIPSRec improves the performance significantly and outperforms state-of-the-art methods.
Conference output Detector Collapse: Backdooring Object Detection to Catastrophic Overload or Blindness in the Physical WorldZhang, H; Hu, S; Wang, Y; Zhang, LY; Zhou, Z; Wang, X; Zhang, Y; Chen, C; Larson, Kate (IJCAI-24: Thirty-Third International Joint Conference on Artificial Intelligence, 2024)Object detection tasks, crucial in safety-critical systems like autonomous driving, focus on pinpointing object locations. These detectors are known to be susceptible to backdoor attacks. However, existing backdoor techniques have primarily been adapted from classification tasks, overlooking deeper vulnerabilities specific to object detection. This paper is dedicated to bridging this gap by introducing Detector Collapse (DC), a brand-new backdoor attack paradigm tailored for object detection. DC is designed to instantly incapacitate detectors (i.e., severely impairing detector's performance and culminating in a denial-of-service). To this end, we develop two innovative attack schemes: Sponge for triggering widespread misidentifications and Blinding for rendering objects invisible. Remarkably, we introduce a novel poisoning strategy exploiting natural objects, enabling DC to act as a practical backdoor in real-world environments. Our experiments on different detectors across several benchmarks show a significant improvement (~10%-60% absolute and ~2-7x relative) in attack efficacy over state-of-the-art attacks.
Conference output FedPFT: Federated Proxy Fine-Tuning of Foundation ModelsPeng, Z; Fan, X; Chen, Y; Wang, Z; Pan, S; Wen, C; Zhang, R; Wang, C; Larson, Kate (IJCAI-24: Thirty-Third International Joint Conference on Artificial Intelligence, 2024)Adapting Foundation Models (FMs) for down- stream tasks through Federated Learning (FL) emerges a promising strategy for protecting data privacy and valuable FMs. Existing methods fine- tune FM by allocating sub-FM to clients in FL, however, leading to suboptimal performance due to insufficient tuning and inevitable error accumula- tions of gradients. In this paper, we propose Feder- ated Proxy Fine-Tuning (FedPFT), a novel method enhancing FMs adaptation in downstream tasks through FL by two key modules. First, the sub-FM construction module employs a layer-wise com- pression approach, facilitating comprehensive FM fine-tuning across all layers by emphasizing those crucial neurons. Second, the sub-FM alignment module conducts a two-step distillations—layer- level and neuron-level—before and during FL fine- tuning respectively, to reduce error of gradient by accurately aligning sub-FM with FM under theo- retical guarantees. Experimental results on seven commonly used datasets (i.e., four text and three vi- sion) demonstrate the superiority of FedPFT. Our code is available at https://github.com/pzp-dzd/FedPFT.
Conference output Gradformer: Graph Transformer with Exponential DecayLiu, C; Yao, Z; Zhan, Y; Ma, X; Pan, S; Hu, W; Larson, Kate (IJCAI-24: Thirty-Third International Joint Conference on Artificial Intelligence, 2024)Graph Transformers (GTs) have demonstrated their advantages across a wide range of tasks. However, the self-attention mechanism in GTs overlooks the graph's inductive biases, particularly biases related to structure, which are crucial for the graph tasks. Although some methods utilize positional encoding and attention bias to model inductive biases, their effectiveness is still suboptimal analytically. Therefore, this paper presents Gradformer, a method innovatively integrating GT with the intrinsic inductive bias by applying an exponential decay mask to the attention matrix. Specifically, the values in the decay mask matrix diminish exponentially, correlating with the decreasing node proximities within the graph structure. This design enables Gradformer to retain its ability to capture information from distant nodes while focusing on the graph's local details. Furthermore, Gradformer introduces a learnable constraint into the decay mask, allowing different attention heads to learn distinct decay masks. Such an design diversifies the attention heads, enabling a more effective assimilation of diverse structural information within the graph. Extensive experiments on various benchmarks demonstrate that Gradformer consistently outperforms the Graph Neural Network and GT baseline models in various graph classification and regression tasks. Additionally, Gradformer has proven to be an effective method for training deep GT models, maintaining or even enhancing accuracy compared to shallow models as the network deepens, in contrast to the significant accuracy drop observed in other GT models. Codes are available at https://github.com/LiuChuang0059/Gradformer.
Conference output Privacy-Preserving Deep Reinforcement Learning based on Differential PrivacyZhao, W; Sang, Y; Xiong, N; Tian, H (2024 International Joint Conference on Neural Networks (IJCNN), 2024)Deep reinforcement learning, with its extensive applications and remarkable performance, is emerging as a pivotal technology garnering researchers' attention. During the training process, there are frequent interaction and data exchange between agents and the environment, and the interaction information during training is closely tied to the training environment. Consequently, this process introduces a high risk of environmental privacy leakage. Malicious third parties may potentially steal state transition matrix or environmental information about the application domain of agent training, resulting in the compromise of user privacy. To address this issue, we propose novel differentially private value-based and policy-based deep reinforcement learning algorithms. Our methods have an advantage of being adaptable to various environmental privacy concerns. We also evaluate them in a customized experimental environment. Comparative experiments are conducted between the original and differentially private versions of the algorithms. The results indicate that our proposed approach can provide differential privacy protection to environmental information with minimal impact on algorithm performance, ultimately achieving a good balance between privacy and utility.
Conference output Robust Learning to Noisy Labels for Semantic Segmentation of Mangrove Communities in Remote Sensing ImageryOtsu, M; Zhou, J (IGARSS 2024 - 2024 IEEE International Geoscience and Remote Sensing Symposium, 2024)The recent focus on automation in remote sensing image analysis has raised attention to robust learning with enormous amounts of data with noisy labels. Nevertheless, research on deep learning with noisy datasets has been rarely designed for semantic segmentation on remote sensing data. To address this issue, we present a case study on a mangrove satellite image dataset with noisy labels and aim to improve pixel-wise classification accuracy and regional coherence. Our method combines data selection and edge-enhancement techniques in noisy data learning. The results demonstrate that the proposed method outperforms the previous data selection method and suggest that this combination is effective in mitigating the negative effects of noisy labels.
Conference output Neuron Efficiency Index: An Empirical Method for Optimizing Parameters in Deep LearningAzam, B; Kuttichira, DP; Verma, B (2024 International Joint Conference on Neural Networks (IJCNN), 2024)Deep Neural Networks (DNNs) have undeniably achieved groundbreaking success across diverse applications. Nevertheless, their complex architectures inherently lead to substantial computational demands and memory prerequisites. To surmount these challenges, this research paper introduces a pioneering approach designed to amplify DNN efficiency via a unique iterative pruning technique Neuron Efficiency Index (NEI), that considers activation frequency of each neuron, class sensitivity and redundancy in the dense layer neurons. The central objective of this method is to curtail the computational burden of the model, all the while ensuring that performance remains intact and enhanced. The proposed technique is used to prune state-of-the-art architectures and comprehensive comparison is presented on benchmark dataset MNIST and CIFAR-10. The evaluation presents that proposed NEI improves the model accuracy while reducing the computations and complexity of the architecture. The work contributes to the field of neural network optimization.