Misinformation spreading over the Internet poses a significant threat to both societies and individuals, necessitating robust and scalable fact-checking that relies on retrieving accurate and trustworthy evidence. Previous methods rely on semantic and social-contextual patterns learned from training data, which limits their generalization to new data distributions. Recently, Retrieval Augmented Generation (RAG) based methods have been proposed to utilize the reasoning capability of LLMs with retrieved grounding evidence documents. However, these methods largely rely on textual similarity for evidence retrieval and struggle to retrieve evidence that captures multi-hop semantic relations within rich document contents. These limitations lead to overlooking subtle factual correlations between the evidence and the claims to be fact-checked during evidence retrieval, thus causing inaccurate veracity predictions.

PM2.5 refers to fine particulate matter less than 2.5 micrometers in diameter. PM2.5 is a common air pollutant. It is capable of entering the respiratory system, and is associated with a variety of health issues such as asthma and other diseases. Pedestrians are at risk of exposure to traffic-related PM2.5 due in part to increased numbers of vehicles in city settings and their associated exhaust fumes - a key contributor to PM2.5. In this paper, we present a framework to minimise PM2.5 exposure for pedestrians by helping them avoid areas with high PM2.5 concentration levels. Specifically we predict the concentration levels through an XGBoost model and background concentration levels from official air quality monitoring stations around Melbourne. We factor in real-time, portable, air quality monitoring devices, weather conditions and real-time traffic flow information. The coefficient of determination (R2), root mean squared error (RMSE) and the mean average error (MAE) for the XGBoost model achieves 0.71, 1.98 and 1.1 respectively. The Dijkstra algorithm is then applied to generate the minimum PM2.5 exposure of routes with alternative routes suggested trading off distance and PM2.5 exposure. Compared with the shortest route, experiments show that PM2.5 exposure can be decreased by 11 - 15% with only a marginal increase in route length.

Abstract In recent years, one of the most popular techniques in the computer vision community has been the deep learning technique. As a data-driven technique, deep model requires enormous amounts of accurately labelled training data, which is often inaccessible in many real-world applications. A data-space solution is Data Augmentation (DA), that can artificially generate new images out of original samples. Image augmentation strategies can vary by dataset, as different data types might require different augmentations to facilitate model training. However, the design of DA policies has been largely decided by the human experts with domain knowledge, which is considered to be highly subjective and error-prone. To mitigate such problem, a novel direction is to automatically learn the image augmentation policies from the given dataset using Automated Data Augmentation (AutoDA) techniques. The goal of AutoDA models is to find the optimal DA policies that can maximize the model performance gains. This survey discusses the underlying reasons of the emergence of AutoDA technology from the perspective of image classification. We identify three key components of a standard AutoDA model: a search space, a search algorithm and an evaluation function. Based on their architecture, we provide a systematic taxonomy of existing image AutoDA approaches. This paper presents the major works in AutoDA field, discussing their pros and cons, and proposing several potential directions for future improvements.

Crowd counting is required for many situations and has historically been undertaken using approximate (manual) estimations and measures. Deep learning allows to improve this situation. Modern crowd counting models are commonly based on pixel-wise density maps using deep convolutional neural networks (CNNs) comprising tens of millions of parameters. These models require high-performance GPUs for training and subsequent usage and inference. As such, these models are difficult to deliver to edge devices that have limited computing resources such as surveillance cameras, mobile phones and Internet of Things (IoT)-type devices. This paper proposes a new method to tackle this issue based on three key components: feature fusion, Bayesian Loss [19] and datasets utilising bounding-box annotations to increase the efficiency of the crowd counting task. Experiments show that the proposed method can not only provide accuracy close to the latest state-of-the-art deep learning models, but support real-time inference in edge devices offering limited computational capacity.

The increasing urbanization of the global population has drawn many researchers’ attention to the field of Intelligent Transportation Systems. Numerous hardware and software technologies have been developed to aid in monitoring and managing the flow of traffic on road networks. As digital cameras become increasingly cheaper and able to produce higher quality images, automated video-based traffic management systems can provide a low cost alternative to conventional (expensive) traffic monitoring systems. In this work we evaluate diverse state-of-the-art deep-learning-based vehicle recognition frameworks on datasets containing surveillance footage of heterogeneous and representative traffic data from Melbourne’s road network. We find that the YOLOv5 family of models offers the optimal balance between detection accuracy, model size, and real-time detection capability for resource-constrained traffic monitoring devices.

To improve social media analysis across diverse platforms, an effective method to evaluate the possibility that different accounts belong to the same users is required. This might be used to support fake news detection or other nefarious activities. In this paper, we present an approach to calculate the probability that different social media accounts on diverse social media platforms belong to the same user. We consider various platform aspects related to user accounts that can be used for user matching including the selected username, the avatar or profile picture, the content of platform posts and related metadata such as the writing style, account binding and use of hyperlinks across platforms. The experimental results showed that the approach is able to distinguish whether the same person has different platform accounts with an F1-score of up to 0.937.

Computer vision opportunities based on deep learning have seen an increasing trend in recent years through the proliferation of convolutional neural networks (CNN) and related deep learning models. However, limited training data often constrains the performance and accuracy of such models - this is a common situation for many detection and classification tasks especially in the ecological field. In this paper, we present a case study of how one such model: You Only Look Once (YOLO) version 5 can be applied to individual feral cat identification using a small unbalanced data set. We describe the procedures for preparing the training and validation data set, training the model using data augmentation and transfer learning techniques, and testing the model on both in-distribution and out-of-distribution samples. We explore the effectiveness of data augmentation methods including basic image manipulations and more advanced augmentation techniques that are now available.

Deepfakes allow for the automatic generation and creation of (fake) video content, e.g. through generative adversarial networks. Deepfake technology is a controversial technology with many wide reaching issues impacting society, e.g. election biasing. Much research has been devoted to developing detection methods to reduce the potential negative impact of deepfakes. Application of neural networks and deep learning is one approach. In this paper, we consider the deepfake detection technologies Xception and MobileNet as two approaches for classification tasks to automatically detect deepfake videos. We utilise training and evaluation datasets from FaceForensics++ comprising four datasets generated using four different and popular deepfake technologies. The results show high accuracy over all datasets with an accuracy varying between 91-98% depending on the deepfake technologies applied. We also developed a voting mechanism that can detect fake videos using the aggregation of all four methods instead of only one.

Understanding the flow of traffic on road networks is a primary focus of road transport authorities. A range of technologies has been applied to measure throughput and potential congestion on the roads. However, such technologies are currently limited in disambiguating the kinds of vehicles on the road network. Certain vehicles types are especially important to distinguish from other traffic, e.g. trucks and trailers. These are larger vehicles that can have a major impact on the roads and surrounding areas, e.g. due to pollution or the excess noise that they can give rise to. In this paper, we present an approach to automatically detect, classify and count the unique classes of trucks and trailers that are found on the road network of Victoria using approaches based on deep learning. In particular, we consider two leading object detection systems: You Only Look Once version 3 (YOLOv3) and Single Shot Multi-Box Detector (SSD). We compare the models and the performance achieved for truck and trailer classification, including dealing with more common and less common truck and trailer types. We measure the efficiency and accuracy of the approaches based on the mean average precision achieved.

The population of Melbourne is growing at over 100,000 every year. This is impacting on all aspects of society: house prices, health, multi-ethnic society, and transport amongst many others. Social media data is a hugely popular data source for research in many domains from understanding urban environments to predicting election results. In this paper we present an approach to identify commuter travel patterns and calculate the average travel time of commuters around Melbourne using social media data from Twitter, Instagram, FourSquare and Flickr. At present there is no other technology or system that captures this information other than by randomly sampling subsets of the population. To achieve this, the social media data was transformed into travel vectors and subsequently filtered in order to facilitate analysis and reduce noise in the data. Travel patterns were then learned using K-means clustering and use of force-directed graphing techniques. Travel pattern identification was based on social media user movement between suburb boundaries in the city of Melbourne. The calculated travel time was compared to the expected travel time by train. A web based platform was developed to allow analysis of the results. The results show that social media can indeed be used to better understand commuting behaviours, although it is highly dependent on the amount of data and especially geo-coded data from individual users.

Deep learning can bring significant improvements to a range of research areas and application domains. Computer vision is one of the key areas that can benefit from deep learning. In particular, object detection and classification can benefit from such approaches. In this work we focus on counting the number of images of individual fruit that can be used by fruit growers to estimate fruit crop yields. To achieve this, we apply two different state of the art convolutional neural networks (CNNs): Faster R-CNN and Single Shot Detection (SSD). CNNs depend on data for training and tuning of the models. In this paper we establish a dataset containing images for a range of fruit types. Using this data, we apply the models to identify the number of fruit in images and the challenges that are encountered. We present the experimental results of applying these approaches and illustrate their performance including the accuracy, time and loss when counting fruit on trees. We consider the future challenges in scaling this work to deal with more complex issues around fruit estimation at scale.

Deep learning is one of the most popular technologies and research areas in machine learning. Convolutional Neural Networks (CNNs) are a typical artificial neural network underpinning deep learning. They have been used in many fields including image recognition, natural language process, and through games such as AlphaGo. A CNN has many advantages such as efficient feature extraction, the simplicity of data format required and the small number of (hyper-)parameters that are required. This paper focuses on a particular application of deep learning: crowd counting. To address this, we apply a Multi-task, Cascaded Convolutional Neural Network (MTCNN). Compared to other models, this model has a good performance and requires a shorter inference time, with shallower network structure and smaller size. In order to demonstrate the value and feasibility of the technology and provide a friendly operating environment for users, the application was realised on both the iOS and Android platforms. A web platform was also developed to visualize the real-time data using a Firebase server.

Prediction of pollution is an increasingly important problem. It can impact individuals and their health, e.g. asthma patients can be greatly affected by air pollution. Traditional air pollution prediction methods have limitations. Machine learning provides one approach that can offer new opportunities for prediction of air pollution. There are however many different machine learning approaches and identifying the best one for the problem at hand is often challenging. In this paper air pollution data, specifically particulate matter of less than 2.5 micrometers (PM2.5) was collected from a variety of web-based resources and following, data cleansing analysed with different machine learning models including linear regression, Artificial Neural Networks and Long Short Term Memory recurrent neural networks. We consider the accuracy and the ability of these different models to predict unhealthy levels of pollution. The advantages and disadvantages of these models are also discussed.

Deep learning provides the ability to train algorithms (models) that can tackle the problems of data classification and prediction based on deriving (learning) knowledge from raw data. Convolutional Neural Networks (CNNs) provides one commonly used approach for image classification and detection. In this work we describe a CNN-based method for detecting dogs in potentially complex images and subsequently consider the identification of the type/breed of dogs. The results achieve nearly 85% accuracy for breed classification for a set of 50 classes of dogs and 64% accuracy for 120 other less common dog types. An iOS application and associated big data processing infrastructure utilizing a variety of GPUs was used to support the image classification algorithms.

People always spend their time at a few important locations for various activities in groups during specific time slots, called place of activity (POA), e.g., resting at home among family members during night and working at office among colleagues during work time. Inferring such places is significant for not only the precise advertising on the commercial aspect but the identifying rallies or meetings among a group of people and tracking of the target individuals on the aspect of public security, e.g., locating and tracking suspected terrorists for anti-terrorist work. However, it is a challenge to map from big location data to places of activity due to the volume and complexity whilst giving rise to privacy concerns, e.g., personally important place mining. In the paper, a method for POA mining on big location data is proposed, named P-PAM, aiming at big data analytics and privacy concerns. We use a clustering algorithm to discover the place of activity, then adopt location entropy as reference of user diversity and take into account temporal variation, to infer place of activity. Further, robust privacy-preserving mechanisms under differential privacy are embedded into clustering results and location entropy evaluation that accesses to raw location data. We demonstrate the utility of our proposed approach with large-scale location datasets derived from geo-referenced social media. The experimental results suggest that the POA mining approach can successfully scale to big data scenarios whilst preserving individual user privacy.

With the explosive growth of location-aware devices and global adoption of social network applications, enormous volumes of spatiotemporal data are being produced. These can be perceived as gazetteers that record frequently visited locations, e.g. shopping malls and museums, and potentially more sensitive locations, e.g. an individual's home/work locations. Density-based clustering approaches are generally used for gazetteer discovery. However, existing clustering solutions are inefficient for big data scenarios and often disregard mobility features derived from trajectories data. Further, automated gazetteer discovery applications may cause privacy concerns. In this paper, we propose a sensitive gazetteer automated discovery approach based on Ω-cluster with robust privacy controls. The approach identifies sensitive gazetteers from massive trajectory data, with location entropy-based filtering used to reduce the number of uninteresting clusters whilst considering mobility features of trajectories. A parallelized solution is implemented to scale across the cloud using memory-oriented data processing solutions based upon Apache Spark. We embed this algorithm in a privacy-preserving mechanism and subsequently release sanitized gazetteers. Through extensive experiments using synthetic and real trajectory datasets from the location based social network (Twitter), we demonstrate the effectiveness and efficiency of our approach.

Unprecedented volumes of location-based information have been produced as a result of the widespread adoption of social network applications and GPS-enabled devices and sensors. Publication of such location data can provide valuable resources for researchers and government agencies in applications ranging from near real-time population-wide health monitoring to planning for future cities. However, such data hold personally identifying information, which gives rise to many privacy issues. There is thus a pressing need for ways to restrict this inherently identifying location-related information, however ideally we would like to preserve the utility of the data. Importantly, any such solution has to be scalable to large population-wide data scenarios. To tackle this, we introduce a novel differentially private hierarchical location sanitization (DPHLS) approach based on the concept “(α, r)-dataset” implemented through a Variable Order Mobility Markov Model (VO3M). We show how this system allows individual locations in personal trajectories to be protected using selection and frequency perturbation mechanisms using the “(α, r)-dataset”, leveraging past (published) location histories to obfuscate the user location in a flexible and controllable manner. The effectiveness and efficiency of the proposed solution is evaluated through the big data experiments that have been carried out using an OpenStack-based Cloud and Apache Sparkbased platform utilising large-scale social media trajectories. The experimental results suggest that the privacy publication algorithm can successfully scale to big data scenarios whilst retaining the utility of the datasets (trajectories) and preserving individual user privacy.

The value that can be extracted from big data greatly motivates organizations to explore data analytics technologies for better decision making and problem solving in a wide range of application domains. Cloud computing greatly eases and benefits big data analytics by offering on-demand and scalable computing infrastructures, platforms, and applications as services. Big data Analytics-as-a-Service (AaaS) platforms aim to deliver data analytics as consumable services in cloud computing environments in a pay as you go model with Service Level Agreement (SLA) guarantees. Resource scheduling for AaaS platforms is significant as big data analytics requires large-scale computing, which can consume huge amounts of resources and incur high resource costs. Our research focuses on proposing automatic and scalable resource scheduling algorithms to maximize the profits for AaaS platforms while delivering AaaS services to users with SLA guarantees on budgets and deadlines to allow timely responses with controllable costs. In this paper, we model and formulate the profit optimization resource scheduling problem and propose an optimization scheduling algorithm that maximizes profits for AaaS platforms and guarantees SLAs for query requests. Experimental evaluations show that the profit optimization scheduling algorithm performs significantly better in cost saving and profit enhancement compared to the state-of-the-art scheduling algorithms.

In this paper, we present the functional capabilities scoping for a novel eResearch infrastructure providing urban researchers with access to datasets and analytical tools. The AURIN portal provides a "lab in a browser" environment, leveraging a complex, loosely-coupled internal architecture and a growing number of federated data sources. Datasets can be "shopped" for, visually explored and analyzed using a growing number of analytical capabilities orchestrated in a workflow environment. While spatial analytical tasks are at the heart of most targeted research disciplines, AURIN aims to reach beyond the scope of traditional GIS and map-based portals. In this paper, we discuss how the functional requirements of AURIN can be realized to enable exploratory and confirmatory data analysis supported by high performance Web based infrastructure.