Monash University Research Portal
Not a member yet
41460 research outputs found
Sort by
Multi-modal deep learning approaches to semantic segmentation of mining footprints with multispectral satellite imagery
Get PDFExisting remote sensing applications in mining are often of limited scope, typically mapping multiple mining land covers for a single mine or only mapping mining extents or a single feature (e.g., tailings dam) for multiple mines across a region. Many of these works have a narrow focus on specific mine land covers rather than encompassing the variety of mining and non-mining land use in a mine site. This study presents a pioneering effort in performing deep learning-based semantic segmentation of 37 mining locations worldwide, representing a range of commodities from gold to coal, using multispectral satellite imagery, to automate mapping of mining and non-mining land covers. Due to the absence of a dedicated training dataset, we crafted a customized multispectral dataset for training and testing deep learning models, leveraging and refining existing datasets in terms of boundaries, shapes, and class labels. We trained and tested multimodal semantic segmentation models, particularly based on U-Net, DeepLabV3+, Feature Pyramid Network (FPN), SegFormer, and IBM-NASA foundational geospatial model (Prithvi) architecture, with a focus on evaluating different model configurations, input band combinations, and the effectiveness of transfer learning. In terms of multimodality, we utilized various image bands, including Red, Green, Blue, and Near Infra-Red (NIR) and Normalized Difference Vegetation Index (NDVI), to determine which combination of inputs yields the most accurate segmentation. Results indicated that among different configurations, FPN with DenseNet-121 backbone, pre-trained on ImageNet, and trained using both RGB and NIR bands, performs the best. We concluded the study with a comprehensive assessment of the model's performance based on climate classification categories and diverse mining commodities. We believe that this work lays a robust foundation for further analysis of the complex relationship between mining projects, communities, and the environment.</p
PhiSiCal-Checkup:A Bayesian framework to validate amino acid conformations within experimental protein structures
Get PDFAs structural biology and drug discovery depend on high-quality protein structures, assessment tools are essential. We describe a new method for validating amino-acid conformations: “PhiSiCal (ΦΨΧal) Checkup.” Twenty new joint probability distributions in the form of statistical mixture models explain the empirical distributions of dihedral angles hω, Φ, Ψ, Χ1, Χ2, . . .iof canonical amino acids in experimental protein structures. Marginal and conditional probability distributions for subsets of dihedral angles are derived from these joint mixture models. Together, these distributions are employed to measure rapidly the information-theoretic “favorability” of any proposed experimental protein structure. The inferred statistical models and measures overcome several shortcomings and afford improvements over the current state of the art in amino-acid conformation verification. Experimental comparisons are made against current protein conformation verification software. In a number of examples, we pick up outliers that are invisible to current methods. We also calculate, as part of verification, the sensitivity of favorability to small changes in a proposed structure accounting for the precision of coordinates. In some cases a near neighbor of a proposed amino-acid conformation may be either less or more favorable. This raises the question, is the current reliance on fixed “thresholds” for validation a good thing? PhiSiCal-Checkup is freely available for online and offline (open-source) use from https://lcb.infotech.monash.edu.au/phisical/checkup.</p
Who is in? Who is out? Exploring primary school students’ sense of belonging using Photovoice
Get PDFThis article describes an exploratory qualitative study undertaken to gain insights into the perceptions of primary-grade students about their sense of belonging (SB) by capturing student voice. The key question that guided the study was what contributes to the student's sense of belonging in the primary schools of Bangladesh. Twenty-nine grade-four students from two purposively selected schools participated in the photovoice study. To represent voice, students took photos of their school in response to the guiding prompts about what, where, and with whom they felt more connected. Group discussions, related to photos taken by students unpacked the stories behind each photo and allowed for the exploration of their sense of belonging. Photo analysis was undertaken at two levels:the denotation level to determine what or who is depicted, and the connotation level that explored what each object or person represents. The study found three factors, i.e., place, people and practice contributing to students’ sense of belonging. Key findings indicate that students felt a strong sense of belonging when they felt safe and supported in school spaces like the playground, library, and classroom; had reciprocal friendships with the same gender peers; and experienced kindness, fairness, leniency, and consideration from their teachers. Conversely, students lacked a sense of belonging when they faced isolation and bullying based on gender, academic performance, disability, and unhygienic places and practices at their school. The study recommends fostering respectful teacher-student relationships, using cooperative peer teaching, improving hygienic practices in schools, enriching school environments, and increasing peer mentoring to nurture students' sense of belonging.</p
Trend-cycle decomposition in the presence of large shocks
Get PDFWe introduce some refinements of the Beveridge-Nelson filter to help address possible distortions from large shocks. We then compare how the Beveridge-Nelson filter and other popular univariate trend-cycle decomposition methods perform given the extreme outliers associated with the Covid recession. Real-time estimates of the output gap based on the Hodrick-Prescott filter are highly unreliable in the years just prior to the pandemic, although the revised estimates during the pandemic are similar to those of the more reliable Beveridge-Nelson filter. The Hamilton filter suffers from base effects that produce a mechanical spike in the estimated output gap exactly two years after the onset of the pandemic, in line with the filter horizon. Given projected data with a simulated Covid-like shock, both the Hodrick-Prescott and Hamilton filters overstate the true reduction in the output gap and fail to capture the implied movements in trend output. The Hodrick-Prescott filter generates a spurious transitory boom just prior to the simulated shock, while the Hamilton filter produces another mechanical spike exactly two years after the simulated shock, as well as an ongoing divergence in forecasted values of the output gap away from zero. Only the Beveridge-Nelson filter correctly forecasts trend and cycle movements when faced with a Covid-like shock
‘Charge what you think you're worth’:a qualitative study exploring the gender pay gap in medicine and the role of price transparency
Get PDFBackground: The gender pay gap in medicine is entrenched and has a negative impact on economic growth, institutional reputation and financial success, recruitment, retention and job satisfaction of female specialists and patient care. It also discourages women from entering specialist fields of medicine. In the Australian unregulated market setting, female specialists are not simply getting paid less, they are choosing to set lower fees than their male counterparts. Aims: We examine how implicit and explicit gender biases affect how fees are set and the potential role of price transparency in addressing the gender pay gap. Methods: We conducted 27 semi-structured interviews with medical specialists recruited via social media and medical society newsletters between June 2021 and March 2022. Interviews were recorded and transcribed verbatim. Data were analysed using thematic analysis. Twenty surgeons and seven anaesthetists, 15 of whom identified as male and 12 as female, participated in this study. The primary outcomes and measures focused on the perspectives of surgeons and anaesthetists regarding fee-setting practices. Results: A combination of contextual and market- and gender-related factors was the source of the biases that determine fee setting. Further, information asymmetry in medicine in Australia underlies current fee-setting practices, exacerbating and entrenching false perceptions about women's skills as surgeons and anaesthetists. Women tend to internalise these biases, self-regulating their behaviours and how they set their fees. Conclusion: The gender pay gap is pervasive. Greater transparency on fees and quality could be explored as a potential solution to reduce pay inequality.</p
Multi-Label Generalized Zero Shot Chest Xray Classification By Combining Image-Text Information With Feature Disentanglement
No full textIn fully supervised learning-based medical image classification, the robustness of a trained model is influenced by its exposure to the range of candidate disease classes. Generalized Zero Shot Learning (GZSL) aims to correctly predict seen and novel unseen classes. Current GZSL approaches have focused mostly on the single-label case. However, it is common for chest X-rays to be labelled with multiple disease classes. We propose a novel multi-modal multi-label GZSL approach that leverages feature disentanglement andmulti-modal information to synthesize features of unseen classes. Disease labels are processed through a pre-trained BioBert model to obtain text embeddings that are used to create a dictionary encoding similarity among different labels. We then use disentangled features and graph aggregation to learn a second dictionary of inter-label similarities. A subsequent clustering step helps to identify representative vectors for each class. The multi-modal multi-label dictionaries and the class representative vectors are used to guide the feature synthesis step, which is the most important component of our pipeline, for generating realistic multi-label disease samples of seen and unseen classes. Our method is benchmarked against multiple competing methods and we outperform all of them based on experiments conducted on the publicly available NIH and CheXpert chest X-ray datasets.</p
Assessing gender bias in the software used in computer science and software engineering education
Get PDFWomen are underrepresented in Computer Science (CS)/ Software Engineering (SE) and other technology related degrees. As undergraduates, they are also less likely to persist with CS/SE studies than men enrolled in those same courses. Gender correlated differences in personal characteristics, behaviour, and preferences mean that course design decisions may introduce unintended bias. To address this issue, we drew inspiration from the GenderMag method. GenderMag uses personas with evidence-based gender differences in problem-solving traits to detect usability issues in software. In this paper we investigate the personal qualities of CS and SE students, and how these influence their CS/SE learning journey. A series of persona development workshops were held to gather an extensive and unique qualitative dataset capturing the prior experiences, preferences, learning styles, motivations, goals, frustrations, and constraints of CS/SE students. Gender differences were used to construct preliminary male and female student personas. These personas were used in cognitive walkthroughs of software applications commonly used in education, and their performance compared to GenderMag's Tim and Abi. While the student personas were less effective and lacked specificity compared to Abi, they were able to identify issues not detectable with GenderMag. Furthermore, the findings show the utility of persona development workshops as a data collection method and introduce a comprehensive list of CS/SE student qualities that may inspire future investigations.</p
Improving the computational efficiency of adaptive audits of IRV elections
Get PDFAWAIRE is one of two extant methods for conducting risk-limiting audits of instant-runoff voting (IRV) elections. In principle AWAIRE can audit IRV contests with any number of candidates, but the original implementation incurred memory and computation costs that grew superexponentially with the number of candidates. This paper improves the algorithmic implementation of AWAIRE in three ways that make it practical to audit IRV contests with 55 candidates, compared to the previous 6 candidates. First, rather than trying from the start to rule out all candidate elimination orders that produce a different winner, the algorithm starts by considering only the final round, testing statistically whether each candidate could have won that round. For those candidates who cannot be ruled out at that stage, it expands to consider earlier and earlier rounds until either it provides strong evidence that the reported winner really won or a full hand count is conducted, revealing who really won. Second, it tests a richer collection of conditions, some of which can rule out many elimination orders at once. Third, it exploits relationships among those conditions, allowing it to abandon testing those that are unlikely to help. We provide real-world examples with up to 36 candidates and synthetic examples with up to 55 candidates, showing how audit sample size depends on the margins and on the tuning parameters. An open-source Python implementation is publicly available.</p
CL-MRI:self-supervised contrastive learning to improve the accuracy of undersampled MRI reconstruction
Get PDFDeep learning (DL) methods have emerged as the state-of-the-art for Magnetic Resonance Imaging (MRI) reconstruction. DL methods typically involve training deep neural networks to take undersampled MRI images as input and transform them into high-quality MRI images through data-driven processes. However, deep learning models often fail with higher levels of undersampling due to the insufficient information in the input, which is crucial for producing high-quality MRI images. Thus, optimizing the information content at the input of a DL reconstruction model could significantly improve reconstruction accuracy. In this paper, we introduce a self-supervised pretraining procedure using contrastive learning to improve the accuracy of undersampled DL MRI reconstruction. We use contrastive learning to transform the MRI image representations into a latent space that maximizes mutual information among different undersampled representations and optimizes the information content at the input of the downstream DL reconstruction models. Our experiments demonstrate improved reconstruction accuracy across a range of acceleration factors and datasets, both quantitatively and qualitatively. Furthermore, our extended experiments validate the proposed framework's robustness under adversarial conditions, such as measurement noise, different k-space sampling patterns, and pathological abnormalities, and also prove the transfer learning capabilities on MRI datasets with completely different anatomy. Additionally, we conducted experiments to visualize and analyze the properties of the proposed MRI contrastive learning latent space. Code available here.</p
Unpacking help-seeking process through multimodal learning analytics:A comparative study of ChatGPT vs Human expert
Get PDFHelp-seeking is an active learning strategy tied to self-regulated learning (SRL), where learners seek assistance when facing challenges. They may seek help from teachers, peers, intelligent tu-tor systems, and more recently, generative artificial intelligence (AI). However, there is limited empirical research on how learners’ help-seeking process differs between generative AI and hu-man experts. To address this, we conducted a lab experiment with 38 university students tasked with essay writing and revising. The students were randomly divided into two groups: one seeking help from ChatGPT (AI Group) and the other from an experienced teacher (HE Group). To examine their help-seeking processes, we used a combination of statistical testing and process mining methods, analyzing multimodal data (e.g., trace data, eye-tracking data, and conversa-tional data). Our results indicated that the AI Group exhibited a nonlinear help-seeking process, such as skipping evaluation, differing significantly from the linear model observed in the HE Group which also aligned with classic help-seeking theory. Detailed analysis revealed that the AI Group asked more operational questions, showing pragmatic help-seeking activities, whereas the HE Group was more proactive in evaluating and processing received feedback. We discussed factors such as social pressure, metacognitive off-loading, and over-reliance on AI in these different help-seeking scenarios. More importantly, this study offers innovative insights and evidence, based on multimodal data, to better understand and scaffold learners learning with generative AI.</p