Abstract Colorectal polyps are precancerous lesions with a high risk of developing into cancer if left untreated. Colonoscopy is the gold standard for detecting and removing these polyps, but it has high miss rates, especially for small and flat polyps. Deep learning methods have been increasingly used to aid in polyp detection, but their performance remains limited when applied to samples captured under unconstrained conditions or when processing images of small and flat polyps. To address this challenge, we propose PolyNeXt, a novel polyp segmentation network that leverages ConvNeXtV2 and global response normalization (GRN) layers. We train PolyNeXt on a combined dataset of samples from Kvasir-SEG and CVC-ClinicDB and evaluate it on four public distinct datasets: ETIS-LaribPolypDB, CVC-ColonDB, CVC-300, and BKAI-IGH NeoPolyp-Small. PolyNeXt achieves state-of-the-art performance on polyps captured in suboptimal conditions, outperforming other methods in terms of Intersection over Union (IoU) and Dice coefficient. Our work demonstrates that PolyNeXt is an effective polyp segmentation network that can improve the accuracy and reliability of detecting polyps captured under challenging circumstances.

Abstract Speech Emotion Recognition (SER) is an advanced technology for developing intuitive and empathetic human-computer interfaces (HCI). While traditional SER systems have achievement a certain degree of succeed in recognising basic emotions from acted speech in a closed environment, real-world applications necessitate the recognition of more complex emotions. This paper presents a systematic review of deep learning approaches in SER from 2019 to the present, following the PRISMA guidelines, with a specific focus on the bridge between basic and complex SER within unimodal (audio-only) and multimodal frameworks. Analysis was done on the landscape of emotion models, datasets, and state-of-the-art (SOTA) model architectures, including CNNs, RNNs, Transformers, and their hybrids. The results reveal that deep learning has improved performance; the following hybrid models improved considerably; however, unimodal models still struggle with the subtle and often overlapping acoustic features of complex emotions. In contrast, multimodal models that leverage complementary information are consistently superior. Nevertheless, challenges remain, such as the over-reliance on a limited range of non-naturalistic datasets, the subjectivity associated with labelling complex emotions, and models not generalising to the variability in the real world. Finally, a conclusion is drawn by offering a strategic roadmap to guide the continuation of research in recognising complex emotions, including the efficient creation of naturalistic, large datasets for future modelling, the development of more advanced techniques for multimodal fusion, and the targeting of unconsidered but available acoustic features to enhance the modelling of the complexity of human emotions.

In an edge-assisted federated learning (FL) system, edge servers aggregate the local models from the clients within their coverage areas to produce intermediate models for the production of the global model. This significantly reduces the communication overhead incurred during the FL process. To accelerate model convergence, FedEdge, the state-of-the-art edge-assisted FL system, trains clients' models in local federations when they wait for the global model in each training round. However, our investigation reveals that it drives the global model towards clients with excessive local training, causing model drifts that undermine model performance for other clients. To tackle this problem, this paper presents Maverick, a new edge-assisted FL system that mitigates model drifts by training personalized local models for clients through contrastive local training. It introduces a model-contrastive loss to facilitate personalized local federated training by driving clients' local models away from the global model and close to their corresponding intermediate models. In addition, Maverick includes anomalous models in contrastive local training as negative samples to accelerate the convergence of clients' local models. Extensive experiments are conducted on three widely-used models trained on three datasets to comprehensively evaluate the performance of Maverick. Compared to state-of-the-art edge-assisted FL systems, Maverick accelerates model convergence by up to 16.2x and improves model accuracy by up to 12.7%.

The rapid advancement of Artificial Intelligence (AI) has transformed various industries, leading to the widespread distribution of AI models and data across intelligent systems. As modern data driven services increasingly integrate distributed knowledge entities, decentralized learning has become a prevalent approach to training AI models. However, this collaborative learning paradigm introduces significant security vulnerabilities and privacy challenges. This paper presents a comprehensive systematic review on private knowledge sharing in distributed learning, analyzing key knowledge components utilized in leading distributed learning architectures. We identify critical vulnerabilities associated with these components and examine defensive strategies to safeguard privacy while mitigating potential adversarial threats. Additionally, we highlight key limitations in knowledge sharing in distributed learning and propose future research directions to enhance security and efficiency in decentralized AI systems.

This study delves into the assessment of diverse machine learning and deep learning models in the context of rainfall prediction in Australia. Utilizing a comprehensive dataset encompassing all states in the country, sourced from the Australian Bureau of Meteorology, various models were evaluated. These encompassed Random Forest, Gradient Boosting, Logistic Regression, K-nearest Neighbors, Gaussian Naive Bayes, Decision Tree, XGBoost, as well as deep learning architectures like LSTM (Long Short-Term Memory) and GRU (Gated Recurrent Unit). The evaluation was based on their respective test scores, unveiling significant disparities in predictive performance. Among the machine learning models, Gradient Boosting exhibited the highest performance with a test score of 0.9315, closely followed by Random Forest at 0.9282. Notably, these scores surpass those reported in existing models. In contrast, Logistic Regression displayed a comparatively lower performance at 0.7711. The deep learning models demonstrated varying degrees of accuracy, with Neural Network achieving a test score of 0.7954, while both RNN architectures, LSTM and GRU, displayed similar performance levels at 0.625. This study furnishes crucial insights into the effectiveness of different machine learning and deep learning models within this domain. It lays a foundation for future investigations, advocating for the exploration of more advanced techniques and feature engineering methodologies to further refine predictive capabilities in this field.

In the field of Computer-Aided Detection (CADx), the use of AI-based algorithms for disease detection in endoscopy images, especially colonoscopy images, is on the rise. However, these algorithms often encounter performance issues due to obstructions like specular reflection, resulting in false positives. This paper presents a novel algorithm specifically designed to tackle the challenges posed by high specular reflection regions in colonoscopy images. The proposed algorithm identifies these regions and applies precise inpainting for restoration. The process entails converting the input image from RGB to HSV color space and focusing on the Saturation (S) component in convex regions detected using a Hessian-based method. This step creates a binary mask that pinpoints areas of specular reflection. The inpainting function then uses this mask to guide the restoration of these identified regions and their borders. To ensure a seamless blend of the restored regions with the background and adjacent pixels, a feathering process is applied to the repaired regions. This enhances both the accuracy and aesthetic coherence of the inpainted images. The performance of our algorithm was rigorously tested on five unique colonoscopy datasets and various endoscopy images from the Kvasir dataset, using an extensive set of evaluation metrics and a comparative analysis with existing methods consistently highlighted the superior performance of our algorithm.

Edge computing (EC) enables low-latency services by pushing computing resources to the network edge. Due to the geographic distribution and limited capacities of edge servers, EC systems face the challenge of edge distributed denial-of-service (DDoS) attacks. Existing systems designed to fight cloud DDoS attacks cannot mitigate edge DDoS attacks effectively due to new attack characteristics. In addition, those systems are typically activated upon detected attacks, which is not always realistic in EC systems. DDoS mitigation needs to be cohesively integrated with workload migration at the edge to ensure timely responses to edge DDoS attacks. In this paper, we present EdgeShield, a novel DDoS mitigation system that leverages edge servers’ computing resources collectively to defend against edge DDoS attacks without the need for attack detection. Aiming to maximize system throughput over time without causing significant service delays, EdgeShield monitors service delays and migrates workloads across an EC system with adaptive mitigation strategies. The experimental results show that EdgeShield significantly outperforms state-of-the-art solutions in both system throughput and service delays.

The skyrocketing growth of mobile apps and mobile devices has significantly fueled the competition among app developers. They have leveraged the app store capabilities to analyse app data and identify app improvement opportunities. Existing research has shown that app developers mostly rely on in-domain (i.e., same domain or same app) data to improve their apps. However, relying on in-domain data results in low diversity and lacks novelty in recommended features. In this work, we present an approach that automatically identifies, classifies and ranks relevant popular features from cross-domain apps for recommendation to any given target app. It includes the following three steps: 1) identify cross-domain apps that are relevant to the target app in terms of their features; 2) filter and group semantically the features of the relevant cross-domain apps that are complementary to the target app; 3) rank and prioritize the complementary cross-domain features (in terms of their domain, app, feature and popularity characteristics) for adoption by the target app’s developers. We have run extensive experiments on 100 target apps from 10 categories over 15,200 cross-domain apps from 31 categories. The experimental results have shown that our approach to identifying, grouping and ranking complementary cross-domain features for recommendation has achieved an accuracy level of over 89%. Our semantic feature grouping technique has also significantly outperformed two existing baseline techniques. The empirical evaluation validates the efficacy of our approach in providing personalised feature recommendation and enhancing app’s user serendipity.

Research Full paper - Computer Science (CS) education is increasingly becoming popular across the globe. With the increased popularity of CS and Information and Communications Technology (ICT) courses, the ability to measure the similarity/distance between units can aid the decision-making process of education providers during learning path recommendations, evaluating unit exemptions/advanced standings, etc. The ability to quantitatively measure the similarity of two units will assist decision-making processes which are otherwise likely to be tedious and time-consuming. Therefore, in this work, we explore how to quantitatively measure the similarity/distance between two CS/ICT units, in an Australian setting. In this study, we utilize data from multiple CS/ICT courses offered by two Australian education providers. The data is generated as a result of the Australian Computer Society's (ACS) course accreditation process during which each unit of a course is mapped into multiple knowledge areas. We measured the similarity between units in terms of their coverage of the knowledge areas using multiple distance measures. Our work is novel in examining a quantitative measure of unit similarity, by introducing a unified approach across education providers that follow the ACS course accreditation process.

Competition in mobile applications (i.e., apps) is becoming more and more intense with the increase in popularity of smart phones and mobile devices. Previous research shows that app developers spent considerable amount of time in exploiting user feedback to improve their apps. However, relying on own user feedback is insufficient for app survival in such competitive environment. It is highly important for an app developer to learn from competitors in order to keep the rank higher or become topper in the store (e.g., Google Play1). In this work, we present an approach to automatically classify and rank popular and un-popular features from the competitor apps. We follow 3 steps- (1) extract features from competitor app updates (i.e., whatsNew) and user feedback (i.e, reviews), (2) filter and group the review features that are relevant to whatsNew features, then, classify whatsNew features to binary classes, such as, popular and unpopular, (3) rank and prioritize those popular and unpopular features from competitors in order to recommend developers to adopt or avoid those features. The ranking of whatsNew features are done based on whatsNew-to-review relevance, user sentiments, and popularity of those features. We conduct extensive experiments on 840 updates and 262000 reviews of 84 different competitor apps of 10 categories. We found encouraging results from the experiments and the empirical evaluation validates the efficacy of our approach, hence, potential usefulness to the developers.

Information visualisation researchers have posited that author-driven narratives will allow information to be conveyed efficiently and argue for the adoption of storytelling techniques in information visualisation. However, there is limited work describing the effects of author-driven narratives in users' comprehension and memorability of visualisations in relation to interactive visualisations. Recommendations for author-driven visualisation stories are largely based on anecdotal reports or research from the arts, and not on studies in information visualisation. To investigate these issues, we carried out a study that compared purely author-driven narratives with interactive visualisations devoid of author narratives, in terms of comprehension and short-term and long-term memorability. We found that the presence of narration in author-driven stories significantly aided the understanding of information but had no significant effect on the long-term recall of information from visualisations.

A geographic information system (GIS) provides basic location-enabled services for many different applications related to navigation, education, and telecommunications. It is a foundation for analysis and visualization. Testing GIS is critical, but challenging due to the difficulty to assess the correctness of GIS outputs, which is called the test oracle problem of software testing. Metamorphic testing alleviates the problem by constructing metamorphic relations (MRs) among multiple inputs and outputs of the program under test. In this article, a semiautomated metamorphic testing (SAMT) method, based on the formal MR model and an improved adaptive random testing algorithm, was proposed to the GIS. To evaluate the performance of our approach, we conducted a case study on a superficial area calculation program, a typical component of GIS. Six kinds of MR construction methods were suggested for the GIS domain program testing. The experimental results show that SAMT can detect the mutations effectively that could solve the test oracle problem efficiently. More importantly, there is no need to manual participation in the testing process, except for the MR construction.

Collaborative learning methods developed with XR technologies are broadly used in different educational fields such as engineering and healthcare. In this survey paper, we discuss different collaborative learning and teaching methods such as trainings and assessment tools from different perspectives. We look at existing research contributions toward collaborative learning systems of healthcare industry and according to the technologies used to develop the collaborative learning systems. The aim of this survey paper is to analyze the available literature, identify the collaborative learning systems which need further developments and discuss the healthcare, categorize which needs more attention, and bring them to the consideration of the future researchers.

Visual data storytelling has gained widespread adoption as a means of communicating information visualisation. This is partly due to the increased interest in data journalism. Besides being engaging, it has been shown to foster better comprehension and memorability of information to target audiences. The visual data story authoring process involves several stages. However, current tools neither consolidate the visual data story creation process nor integrate the essential features, such as the recommendation of logically sequenced story pieces, for producing coherent narratives. This paper briefly demonstrates our approach and framework for supporting the creation of logically sequenced visual data stories.

Summary Software vulnerability is a critical issue in the realm of cyber security. In terms of techniques, machine learning (ML) has been successfully used in many real‐world problems such as software vulnerability detection, malware detection and function recognition, for high‐quality feature representation learning. In this paper, we propose a performance evaluation study on ML based solutions for software vulnerability detection, conducting three experiments: machine learning‐based techniques for software vulnerability detection based on the scenario of single type of vulnerability and multiple types of vulnerabilities per dataset; machine learning‐based techniques for cross‐project software vulnerability detection; and software vulnerability detection when facing the class imbalance problem with varying imbalance ratios. Experimental results show that it is possible to employ software vulnerability detection based on ML techniques. However, ML‐based techniques suffer poor performance on both cross‐project and class imbalance problem in software vulnerability detection.

Demand in Global Software Engineering (GSE) is increasing every year. While GSE reduces development cost and provides access to resources poll, GSE practitioners also need to deal with many challenges. This impacts Requirements Engineering (RE) process in term of teamwork, collaboration, knowledge management, time and cultural differences. Providing that RE is considered to be an important part of the software development, several studies have pointed out the need of RE process that supports GSE environment. We recognize the importance of RE tools in supporting RE process and conduct the study to discover the best way to use RE tools to solve the challenges in GSE. The study adopts the Feature Analysis Screening Mode approach and generates the list of features with four categories that could address the challenges: (1) Shared Knowledge Management, (2) Workflow and Change Management, (3) Traceability, and (4) System and Data Integration. Four RE tools on the market are selected for inv estigation. We have found out how the tools best support three of the categories but have limited capability for the first category. Some suggestions are provided for the future development to provide the support for RE work in GSE environment.

Effective visualisation plays a vital role in generating insights from data. The selection of graph types however, is highly dependent on the analysis tasks and data types at hand. For example, spatio-temporal visualisations encode changes in data over time and space. Although they have the potential of revealing overall tendencies and movement patterns, building effective spatio-temporal visualisations is challenging because it requires encoding all three attributes of spatio-temporal data i.e. thematic (values of attributes), temporal and spatial in a single visualisation. In this application design study, we present PedaViz for representing hour-level spatio-temporal attributes within a single visualisation; a 24-hour radial visual metaphor that encodes hour-level temporal and daily temperature attributes while utilising a thematic map display to present spatial attributes. The design was applied on city planning domain using Melbourne's pedestrian count and temperature data. Results of our preliminary user evaluation suggest that our visualisation is easily understandable by users; and supports users in carrying out selected analysis tasks.

Learning HTML poses similar challenges as learning conventional programming language by novice programmer. Apart from the HTML validator, there are limited tools to help novice programmer address errors in their HTML code. In this study, we employ visualisation techniques to display the structural and contextual information of the HTML code. We look at condensing and visually representing the important aspects of the HTML code. This is to enable novice programmers gain insights on the HTML code structure and locate any underlying syntax and semantic errors.

We present a visualisation of Melbourne pedestrian count data and a visual metaphor for representing hour-level temporal dimension in this context. The pedestrian count data is captured from sensors located around the city. A visualisation web application is implemented that incorporates a thematic map of these sensor locations with a 24-hour clocklike polygon that shows pedestrian counts at every hour, and alongside a display of daily temperature. Our visualisation allows users to analyse how the city is used by pedestrians. Moreover, the design of our visualisation was driven by the type of analysis tasks carried out by city planners. The visualisation would help city planners better understand the dynamics of pedestrian activity within the city and aid them in urban management and design policy recommendation.

When supervising software engineering team projects, having all team members contribute actively to the project is often a challenge. Most often than not, there will be teams having some members with limited or no contribution. Thus one of the key roles of a team leader and academic supervisor are to monitor who is contributing and who is falling behind. Assessing the progress information of each team member becomes vital. This is to introduce strategies that encourages ensure every member is contributing effectively and efficiently in a timely manner. This paper proposes a rubric solution that assesses the progress information of each team member and provides a formative performance feedback on how each member is contributing to the project. This will enable each team member to reflect on his or her performance, and hopefully self-regulate and put in the necessary contribution. At the same time, this will assist the team leader and academic supervisor in monitoring the team member.

Currently PHP is the most widely used web programming language for websites [1]; however there seems to be a limited availability of debugging tools that a novice programmer can use. In this work, we propose a framework to identify logic errors committed by novice PHP programmers and prototype application to automate the process of detecting such errors. The aim is to develop a tool that may be used to assist novice programmers in their learning process, and contribute to computer science education research.