Monash University Research Portal
Not a member yet
41460 research outputs found
Sort by
Thumb:A forensic automation framework leveraging MLLMs and OCR on Android device
No full textThe forensic of Android devices is challenging due to automated thumbnail generation by applications and the operating system, complicating attribution to specific user actions. This paper presents the design, implementation, and evaluation of a forensic framework, Thumb, which performs real-time experiments on physical Android devices. Thumb integrates multimodal large language models (MLLM) and Optical Character Recognition (OCR) to capture on-screen information and simulate user interactions, while extracting data from internal storage to monitor changes in cached and thumbnail files. A proof-of-concept implementation demonstrates the framework's accuracy across various applications, highlighting its potential to simplify Android forensic analysis. However, current MLLM limitations and the framework's structure pose challenges in complex scenarios and detailed data analysis.</p
Contrastive Learning-Based Multi-Level Knowledge Distillation
No full textWith the increasing constraints of hardware devices, there is a growing demand for compact models to be deployed on device endpoints. Knowledge distillation, a widely used technique for model compression and knowledge transfer, has gained significant attention in recent years. However, traditional distillation approaches compare the knowledge of individual samples indirectly through class prototypes overlooking the structural relationships between samples. Although recent distillation methods based on contrastive learning can capture relational knowledge, their relational constraints often distort the positional information of the samples leading to compromised performance in the distilled model. To address these challenges and further enhance the performance of compact models, we propose a novel approach, termed contrastive learning-based multi-level knowledge distillation (CLMKD). The CLMKD framework introduces three key modules: class-guided contrastive distillation, gradient relation contrastive distillation, and semantic similarity distillation. These modules are effectively integrated into a unified framework to extract feature knowledge from multiple levels, capturing not only the representational consistency of individual samples but also their higher-order structure and semantic similarity. We evaluate the proposed CLMKD method on multiple image classification datasets and the results demonstrate its superior performance compared to state-of-the-art knowledge distillation methods.</p
Screentone-Preserved Manga Retargeting
No full textAs a popular comic style, manga offers a unique impression by utilizing a rich set ofbitonal patterns, or screentones, for illustration. However, screentones can easily be degraded when manga is resized in terms of aspect ratio and resolution for manga re-layout and e-manga migration applications. To tackle this problem, we propose the first automatic manga retargeting method that synthesizes a retargeted manga image while preserving the prominent structure and fine screentone intended by the manga artist. While modern natural photo retargeting methods can achieve prominent structure preservation, preserving screentones within arbitrarily shaped regions is very challenging due to two properties of manga: (i) pattern constancy under translation, and (ii) non-compatibility with interpolation. To circumvent this barrier, we propose learning a quantized representation of screentones that is translation-invariant and pointwisely representable through a tailored manga reconstruction network with a screentone-anchored codebook. Thanks to these merits, we can perform the re-synthesis operation using existing photo retargeting methods and achieve the desired manga retargeting results. We conducted extensive qualitative and quantitative experiments to validate the effectiveness of our method, and we achieved notably compelling results compared to alternative methods.</p
Pride and persistence:social comparisons in production
No full textWork is ordinary and necessary for most people, but some people work excessively (“work persistence”), seemingly driven by internal forces. We theoretically and experimentally investigate the role of relative performance incentives in causing or exacerbating work persistence. In our setting, agents perform a task over two stages. In the first stage, they can earn prizes, which are allocated either randomly or according to relative performance. Afterwards, they have the opportunity to continue working in a second stage, with payment by piece rate and no competition against others. Our theoretical model of motivated belief updating predicts that agents adjust their beliefs asymmetrically: they attribute their relative performance more to their productivity if they win a prize, and more to luck if they lose. This bias leads winners of the first-stage prize to increase their effort in the subsequent piece-rate stage, but with no corresponding decrease in work effort by losers. Results from a real-effort experiment confirm these predictions: winners' effort in the piece-rate stage is roughly 30 percent higher when earlier bonus prizes had been allocated by performance, compared to when those prizes had been allocated randomly. Losers' effort is also higher – not lower – though this difference is not significant.</p
Are auditors insulated to positive client news? Evidence from audit fees and going-concern opinions
No full textThis paper investigates if and how auditors consider positive news, related to their clients, when making audit decisions. Using a large sample of US listed companies, we find that auditors charge lower audit fees and are less likely to issue going-concern opinions when there are a higher number of positive news items related to their clients. Additional analyses show that the influence of positive client news on audit decisions varies with news characteristics, client size, auditor tenure, and auditor size. We also find some evidence that auditors' reactions to positive client news varies with the topic of news, and positive client news has implications for financial reporting quality and accuracy of going-concern opinions. Collectively, these findings suggest that the media affects auditors' business risk and that auditors are not insulated to positive news.</p
Substance use prevalence and the convergent validity of rating scales in a child and youth mental health service
No full textObjectives: Although some decline in substance use among young Australians is evident, substance use disorder (SUD) continues to compromise mental health outcomes. The current study examined the prevalence of substance use problems and SUD among young people attending a child and youth mental health service. The convergent validity of three routine clinical rating scales with a SUD diagnosis, and with each other, was also investigated. Method: Deidentified data from the Health of the Nation Outcome Scales for Children and Adolescents (HoNOSCA), the Alcohol, Smoking and Substance Involvement Screening Test (ASSIST), and the Clinical Risk Assessment and Management (CRAM) checklist were extracted from the clinical files of 90 young people attending the service. Results: Thirty-three percent of young people had a substance use problem and 14% met SUD criteria in addition to their mental health diagnosis. Iterative convergent validity analyses demonstrated HoNOSCA and CRAM substance use problem ratings aligned closely with a diagnosed SUD. ASSIST ratings differed significantly from HoNOSCA, CRAM, and a diagnosed SUD. Conclusions: The findings support the value of HoNOSCA in detecting substance use amongst young people presenting at a child and youth mental health service. ASSIST had lower completion rates suggesting lower clinical utility.</p
The future of paediatric sleep medicine:a blueprint for advancing the field
No full textPaediatric sleep medicine has rapidly evolved and expanded over the past half century as it became increasingly recognised as a unique field related to but distinct from adult sleep medicine. In looking forward to the next years, the focus of the following discussion is two-fold: to summarise a brief history of the field, recent developments and current trends, and to present a blueprint for the future across various key domains. Using Bronfenbrenner's Ecological Systems Theory as a model for the interaction between the five interconnected ecosystems and sleep in children, we discuss a variety of topics relevant for the present state and future of paediatric sleep medicine. Such topics include the potential effects of climate change and war on children's sleep, the development of public policy initiatives—such as sleep education in schools and in communities, and global efforts to reduce the epidemic of insufficient sleep. Indeed, insufficient sleep contributes to a myriad of negative medical, mental health, functional, and safety consequences. We also focus on the development of paediatric sleep medicine-specific educational initiatives and training programmes, and we showcase professional organisations such as the International Paediatric Sleep Association that are dedicated to the global expansion of paediatric sleep medicine. Finally, we address the need for further interdisciplinary collaborations, identify critical research gaps and explore the potential role of artificial intelligence and other new technologies in paediatric sleep research, including standardisation of sleep measurements, and novel methods of monitoring sleep in children.</p
Language diversity and teaching practices:Japanese high school teachers of English
No full textIn Japan, as in many Asian countries, English is still widely taught in schools in standardised forms that are attributed to ‘native speakers’. However, scholarship on linguistic diversity is well-established in the field of Teaching English to Speakers of Other Languages (TESOL), and teachers’ conceptualisation of language can influence their teaching in different ways. In this article we discuss the relationship between six Japanese high school teachers’ understanding of English and their reported teaching practices. Data collected via individual interviews and email exchanges were analysed, and the findings revealed various conceptualisations of language diversity, ranging from a focus on Inner Circle varieties to World Englishes, English as a ‘tool’ for communication, and translanguaging. These ways of thinking about language were found to align with different kinds of reported classroom activities.</p
Humanism strikes back? A posthumanist reckoning with ‘self-development’ and generative AI
No full textSince the release of OpenAI's ChatGPT in 2022, AI activity has reached a fever pitch. Calls for effective ethical responses to the pressurised AI environment have in turn abounded. Posthumanism, which seeks to build ethical futures by de-centring the ‘human’, is an obvious candidate to act as a lynchpin of theoretical intervention. In their responses, posthumanist scholars appear to have embraced AI’s potential to destabilise Humanist philosophical ideas. We critically interrogate this initial enthusiasm. Conceptually distinguishing ‘post-dualist self-development’ (PDSD) from ‘technical self-development’ (TSD), we show how AI prompts an urgent need to advance posthumanist engagement with how technical development unsupervised by humans is ontologically discrete from other forms of material agency. We argue that specific engagement with TSD as distinct from PDSD is a key to avoid ignoring or underestimating Humanist and anthropocentric aspects of current AI innovation, and the influence of anthropomorphism. Without a theoretical reckoning with these tensions, posthumanism in the AI-era runs the risk of potentially promoting technologies that reinvigorate Humanist and anthropocentric expansion. To conclude, we show how a posthumanist ethics of generative AI that pays requisite attention to both TSD and PDSD may enable more anticipatory and nuanced assessments of the risks and benefits of discrete AI technologies to inform public discourse, appropriate social, institutional, policy and governance responses, and direct AI research and development priorities.</p
Do brokers manage the distribution of stock recommendations?
No full textThis study examines whether and how brokers manage the distribution of their stock recommendations. We document that if a broker's percentage of buy recommendations in a quarter is substantially higher than its target level, the broker issues significantly fewer buy recommendations than other brokers in the following quarter. This evidence remains robust after controlling for mean reversion in the data and varies systematically with brokers' expected benefits and costs of managing the distribution. Exploring possible methods to manage the distribution, we find evidence suggesting that brokers alter the timing of recommendation initiations and reiterations, and shift recommendations between adjacent quarters. Finally, we show that distribution management affects the informativeness of stock recommendations in the market.</p