To address the growing threats of Zero-Day attacks, we propose an advanced intrusion detection system framework that integrates a GAMO model for data balancing and a Double DQN mechanism for dynamic sample selection. Unlike existing methods relying on static thresholds, our framework continuously adjusts its sampling strategies based on changing threat landscapes, thereby enhancing Zero-Day attack detection and improving the recognition of minority classes in imbalanced datasets. We validate the proposed framework using the CICIDS2017 dataset. In binary classification experiments, our approach achieves 99.66% accuracy, outperforming multiple baseline models. The GAMO component enhances the detection rates for minority attack classes.

Pre-trained vision-language models (VLMs) exhibit significant vulnerability to imperceptible adversarial perturbations. Current advanced defense strategies typically employ adversarial prompt tuning to improve the adversarial robustness of VLMs, which struggle to simultaneously maintain generalization across both natural and adversarial examples under different benchmarks and downstream tasks. We propose a collaborative adversarial prompt tuning (CoAPT) approach from pre-trained VLMs to target robust VLMs. Inspired by the image mask modeling, we adopt an improved real-time total variation algorithm to suppress and eliminate high-frequency details from images while preserving edge structures, thereby disrupting the adversarial perturbation space. Subsequently, guided by the high-level image and text representations in the latent space of the pre-trained VLMs, the corrupted natural features are restored while inheriting the superior generalization capability. Experiments on four benchmarks demonstrate that CoAPT achieves an excellent trade-off among natural generalization, adversarial robustness, and task-specific adaptation compared to state-of-the-art methods.

As networks become more complex and interconnected, they grow more vulnerable to sophisticated cyberattacks. CyberBattleSim provides a simulation platform for modeling complex attack-defense interactions. This paper presents a real-time defense framework based on deep reinforcement learning, integrating Proximal Policy Optimization (PPO) and Asynchronous Advantage Actor-Critic (A3C). While prior research has primarily focused on improving attacker strategies such as Rapid ActorCritic (RAC) and Double Deep Q-Learning (DDQL), this work shifts attention to defender adaptability. The proposed Parallel Proximal Policy Optimization (PPPO) framework combines PPO’s policy stability with A3C’s asynchronous parallelism to enable rapid adaptation to evolving threats. Simulation results demonstrate that PPPO outperforms baseline approaches, including Deep Q-Learning and random policies in terms of cumulative rewards and network availability. Notably, PPPO maintains high availability even under multiple attacker scenarios. Real-time evaluations reveal a peak response latency of 0.005 seconds when utilizing five parallel agents. Comparative experiments further highlight the efficiency of GPU-based implementations, which achieve significantly faster training than CPU-based versions. These findings underscore the effectiveness and scalability of PPPO in enabling adaptive, autonomous defense mechanisms. This study contributes to advancing deep reinforcement learning applications in cybersecurity by providing a robust and real-time defense strategy capable of mitigating unknown and dynamic attacks in complex network environments.

ABSTRACT Federated learning (FL) enables collaborative model training across multiple parties while preserving data privacy. However, FL remains vulnerable to privacy leakage through model updates. Differential privacy (DP) has been incorporated into FL by adding noise to model updates to ensure robust privacy protections. Traditional DP methods set a fixed sensitivity limit, resulting in excessive noise addition and performance degradation, especially in complex models such as RS, where the high‐dimensional parameters of the model pose a major challenge to effective noise addition. The higher the parameter dimension of the model, the greater the effect of including all the noise will be on the performance of the recommendation system. This study presents an adaptive local differential privacy technique grounded in Fisher information for reinforcement learning, which dynamically modifies the privacy budget by assessing the significance of model parameters during each training iteration. Specifically, Fisher information is used to assess the significance of each parameter layer, with more noise added to less important layers and less noise added to more critical layers. This approach optimizes the noise allocation while maintaining model performance under the DP guarantee. At the same time, we decouple the federated recommendation system (FRS) from the DP mechanism, enabling seamless integration with a variety of recommendation models. We assess the effectiveness of the proposed method through theoretical analysis and experiments on multiple benchmark datasets. The results demonstrate that the Fisher‐based adaptive DP method significantly improves model performance compared to traditional fixed‐sensitivity DP methods in FL environments, particularly addressing the challenges posed by the complexity of RS parameters.

Abstract In the western world,the prostate cancer is major cause of death in males. Magnetic Resonance Imaging (MRI) is widely used for the detection of prostate cancer due to which it is an open area of research. The proposed method uses deep learning framework for the detection of prostate cancer using the concept of Gleason grading of the historical images. A3D convolutional neural network has been used to observe the affected region and predicting the affected region with the help of Epithelial and the Gleason grading network. The proposed model has performed the state-of-art while detecting epithelial and the Gleason score simultaneously. The performance has been measured by considering all the slices of MRI, volumes of MRI with the test fold, and segmenting prostate cancer with help of Endorectal Coil for collecting the images of MRI of the prostate 3D CNN network. Experimentally, it was observed that the proposed deep learning approach has achieved overall specificity of 85% with an accuracy of 87% and sensitivity 89% over the patient-level for the different targeted MRI images of the challenge of the SPIE-AAPM-NCI Prostate dataset.

Nowadays, means of communication among people have changed due to advancements in information technology and the rise of online multi-social media. Many people express their feelings, ideas, and emotions on social media sites such as Instagram, Twitter, Gab, Reddit, Facebook, and YouTube. However, people have misused social media to send hateful messages to specific individuals or groups to create chaos. For various governance authorities, manually identifying hate speech on various social media platforms is a difficult task to avoid such chaos. In this study, a hybrid deep-learning model, where bidirectional long short-term memory (BiLSTM) and convolutional neural network (CNN) are used to classify hate speech in textual data, is proposed. This model incorporates a GLOVE-based word embedding approach, dropout, L2 regularization, and global max pooling to get impressive results. Further, the proposed BiLSTM-CNN model has been evaluated on various datasets to achieve state-of-the-art performance that is superior to the traditional and existing machine learning methods in terms of accuracy, precision, recall, and F1-score.

With a great percentage of farms in New Zealand as pastures, they are mainly important in contributing to the milk and meat industries. Pasture quality is highly affected by weeds. Weeds grow fast and invade pastures by seed pollination. They consume the nutrients, water, and other minerals, and once they are bitter, cattle do not eat them. Therefore, dairy farmers have to allocate a significant portion of their budget and time to monitor and clean weeds. Unfortunately, most weed management tasks are manual with no consistent technology. Thus, the motivation behind this article was to design an object detection model for weed monitoring and control in pastures. The model was designed and tested on California thistle, a dominant and widespread weed on New Zealand pastures. Our study is one of the major model designs for identifying weeds in an in-pasture environment, one of the most complicated environments for any object detection model. A synthetic methodology was used to create three types of datasets: plant-based, leaf-based, and mixed. The trained model based on the leaf-based dataset is one of the major contributions of our work and has not been conducted by any other weed detection models. After models had been trained, tuning experimentation was undertaken to improve the model’s performance. This involved studying the model’s hyperparameters in various ranges and then recording their values at the optimum points. The improved model showed a 93% mAP accuracy in the detection of training images and over 95% accuracy for testing images. The experimentation showed that the leaf-based model was slightly better than other models. The model can automate highly any weed management system. The use of this model will save farmers time and money and help them reduce the errors of manual work.

Cervical cancer is one of the common cancers among women and it causes significant mortality in many developing countries. Diagnosis of cervical lesions is done using pap smear test or visual inspection using acetic acid (staining). Digital colposcopy, an inexpensive methodology, provides painless and efficient screening results. Therefore, automating cervical cancer screening using colposcopy images will be highly useful in saving many lives. Nowadays, many automation techniques using computer vision and machine learning in cervical screening gained attention, paving the way for diagnosing cervical cancer. However, most of the methods rely entirely on the annotation of cervical spotting and segmentation. This paper aims to introduce the Faster Small-Object Detection Neural Networks (FSOD-GAN) to address the cervical screening and diagnosis of cervical cancer and the type of cancer using digital colposcopy images. The proposed approach automatically detects the cervical spot using Faster Region-Based Convolutional Neural Network (FR-CNN) and performs the hierarchical multiclass classification of three types of cervical cancer lesions. Experimentation was done with colposcopy data collected from available open sources consisting of 1,993 patients with three cervical categories, and the proposed approach shows 99% accuracy in diagnosing the stages of cervical cancer.

In the file sharing ecosystem, One-Click File Hosting Services (FHS) such as Rapidgator and Uploaded, the previously Rapidshare and Megaupload, provide a platform for users to share copyrighted content. We present a publisher-side analysis of FHS file sharing dynamics through data collected from active measurement by crawling Warez-BB. The website is essentially a forum where publishers can share links to content they have uploaded on file hosting services. Consumers can use the website to gain access to content shared on the website, often free of charge. We primarily analyse various characteristics of file sharing with respect to view count as the evaluation metric.

This paper aims to develop a real-time Plant Environment Simulator (PES), which simulates a corrugated plant effectively and realistically. The resultant solution of this work can be used to provide factory workers or new developers with a responsive, simulated learning environment on teaching how to use existing software correctly. The work is carried out for a large cardbox maker that can be used to test new prototypes without using the actual plant facilities, so it will economically and efficiently contribute to the creation of new robust software products for the corrugated plant.

Cultural venues like museums increasingly seek to harness the value of data analytics to make data driven decisions related to exhibitions duration, marketing campaigns, resource planning, and revenue optimization. One key priority is the need to understand the influencing factors behind visitor attendance. Using data collected from a large museum, we investigated whether the weather has a significant impact on visitor attendance or that other factors are more important. We applied the Cross Industry Standard Process for Data Mining (CRISP-DM) methodology to perform the research, developed and built four different types of regression models using R and its machine learning packages to model visitor attendance. The models were trained and evaluated. Predictions of visitor attendance were then generated from each of the four models and forecast accuracy was measured. The extreme gradient boost model was the best model with the highest average forecast accuracy of 93% and lowest forecast variability when benchmarked against the actual visitor attendance from the test data set. The weather was not considered to be as significant in predicting visitor trends and numbers to the museum compared to factors like time of the day, day of the week and school holidays. However, it was still measured to have a slight impact as excluding weather variables resulted in a model with a poorer fit. Weather can potentially have a more marked impact on cultural attractions in more extreme weather environments and outdoor venues.

Pornography can be distributed in multiple forms on the Internet. Online pornography forms a non-negligible fraction of the total Internet traffic, with adult video streaming gaining significant traction among the most visited global websites. Similar to the rise of User Generated Content (UGC) on general Web 2.0 services, adult video service providers have also promoted social interaction and UGC in what is called 'Porn 2.0'. Discovering the characteristics of Porn 2.0 allows for better understanding of both Internet traffic in general and specifically UGC services. In this paper, using trace-driven analysis, we examined the characteristics of one of the most well-known Porn 2.0 service providers, XHamster. We found that a large proportion of the currently available videos were uploaded in recent years and this has coincided with a rapid growth in the use of video categories. Compared to non-adult UGC services, we found user interaction on XHamster to revolve more strongly around ratings than comments and the average duration and views per video were higher.

Internet piracy is a significant ongoing problem for content producers and rights holders. Estimates of the cost of copyright infringement to the film and music industries range in the tens to hundreds of billions of dollars every year. The vast majority of this illegal content is shared using three key technologies: peer-to-peer (P2P) protocols such as BitTorrent, illegal file streaming, and one-click file hosting services (OCHs). The current dominant analogy for file- sharing, promoted by copyright holders and industry lobby groups, is one of 'copyright theft'; with content uploaders predominantly depicted as opportunists motivated by financial gain. Recently, academics from various disciplines have begun to question this narrative, proposing alternative models for understanding piracy based on the concept of the social or 'altruistic' sharer. In this paper, two OCH indexes were studied for insights into uploader dynamics. Results suggest that traditional understandings of Internet piracy are significantly limited in their ability to explain a number of aspects of the current OCH ecosystem. A significant number of uploaders are found to be behaving in ways that do not fit the traditional economic narrative; large numbers of users are making negligible money, and aggregate figures show a significant amount of uploaders are failing to take actions to appropriately maximise their hypothetical earnings.

Adult video content is considered to be responsible for a major portion of the Internet traffic. We present a preliminary characterization study of one of the largest and popular adult streaming service, xHamster. While there have been extensive analysis of mainstream user generated content sharing services such as YouTube, adult streaming services have been an underappreciated field of study for researchers. Using metadata of over 1.8 million videos we studied the video characteristics, user interaction, and popularity dynamics of xHamster. We observe that adult services garner more views per video than YouTube. Adult streaming sites tend to have longer videos compared to mainstream streaming sites. In adult streaming services it is common for videos to be assigned multiple categories.

Internet piracy is a significant ongoing problem for content producers and rights holders. Estimates of the cost of copyright infringement to the film and music industries range in the tens to hundreds of billions of dollars every year. The vast majority of this illegal content is shared using three key technologies: peer-to-peer (P2P) protocols such as BitTorrent, illegal file streaming, and one-click file hosting services (OCHs). In this paper, two OCH indexes were studied for insights into uploader dynamics. Results suggest that traditional understandings of internet piracy are significantly limited in their ability to explain a number of aspects of the current OCH ecosystem. A significant number of uploaders are found to be behaving in ways that do not fit the traditional economic narrative.

This paper presents the design, implementation, and evaluation of an energy-efficient handover algorithm for the wireless sensor networks (WSNs) that are the main building block in the creation of the Internet of Things (IoT). Our low-power handover design is based on a careful breakdown and analysis of the potential power consumption of different components of the handover process. With the scanning part of the process being identified as the main drain of energy, the algorithm is designed to place the majority of the scanning responsibility on the mains powered access points, rather than on the low-power mobile nodes. The proposed algorithm has been implemented and its functionality and low power consumption have been empirically evaluated. We show that the design can reduce the energy consumption by several orders of magnitude compared to existing handover algorithms for WSNs. These reductions substantially extend the lifetime of low-power IoT devices with fixed battery capacity and reduce their battery requirements of other IoT devices.

Measuring quality of Internet access is important because it provides rich information on capability of Internet services and user experience. As Internet services actively evolve, users tend to spend more time and require more network capacity over time. To meet the needs of consumers, there is a wide mix of ISPs and access technologies offered in New Zealand. Benchmarking ISPs in New Zealand for their quality of service enables us to predict what users may experience among different ISPs. In this paper, we present a study of broadband service performance in New Zealand ISPs. Our results will provide accurate information on the quality of service users experience from New Zealand ISPs and helps both ISPs and Government to understand current Internet service market.

When understanding modern web usage and providing optimized personalized service, it is important to identify the HTTP(S) requests directly caused by user actions like clicks and typing web addresses. With a majority of HTTP(S) requests being due to content that has not been explicitly requested by a user, the problem of identifying user actions at proxies or middleboxes becomes non-trivial. We present an automated evaluation framework for identifying user actions while also automatically providing a "ground truth" of the user actions. We utilize the framework to compare the performance of timing-based and HTTP-aware request classifiers, including timing-based classifiers operating on both per-request and per-connection basis to identify user actions. We emphasize the value of diverse information used by the classifiers when comparing identification accuracy both among classifiers and relative to the browser-based ground truth. Our classifiers can be useful to better understand users' web usage and connection prioritization.

Large-scale scientific installations generate voluminous amounts of data (or big data) every day. These data often need to be transferred using high-speed links (typically with 10 Gb/s or more link capacity) to researchers located around the globe for storage and analysis. Efficiently transferring big data across countries or continents requires specialized big data transfer protocols. Several big data transfer protocols have been proposed in the literature, however, a comparative analysis of these protocols over a long distance international network is lacking in the literature. We present a comparative performance and fairness study of three open-source big data transfer protocols, namely, GridFTP, FDT, and UDT, using a 10 Gb/s high-speed link between New Zealand and Sweden. We find that there is limited performance difference between GridFTP and FDT. GridFTP is stable in terms of handling file system and TCP socket buffer. UDT has an implementation issue that limits its performance. FDT has issues with small buffer size limiting its performance, however, this problem is overcome by using multiple flows. Our work indicates that faster file systems and larger TCP socket buffers in both the operating system and application are useful in improving data transfer rates.

This paper presents a characterization study of big data transfer protocols on a long-haul network. We analyzed the performance and fairness of three well-known open-source protocols, namely, GridFTP, FDT, and UDT. Using a real-world 10 Gb/s network link between New Zealand and Sweden, we studied data transfer rates (in terms of goodput) and fairness (in terms of impact on round trip time) of the protocols. We performed extensive experiments using single and multiple data flows to comprehend how these protocols behave in real-world situations. We found that GridFTP has the fastest data transfer rates when using a single flow. UDT suffered from poor performance due to implementation issues. A small buffer size limited FDT's performance, however, this drawback can be overcome by using multiple flows in lieu of fairness.

Researchers working in diverse fields such as astronomy, experimental physics, genomics, and meteorology have to frequently deal with analyzing voluminous amounts of complex data. Such data is often referred to as big data. These researchers work in teams and have to transfer this data over long distances. Efficiently transferring big data over long distances requires the use of appropriate transfer protocols. Several TCP-based and UDP-based protocols have been proposed in the literature, however, a comparative analysis of such protocols is lacking in the literature. This paper presents a comparative performance analysis of four well-known high-speed data transfer protocols for long fat networks, namely, GridFTP, FDT, UDT, and Tsunami. We performed extensive experiments to measure the effectiveness of each protocol in terms of its throughput for various roundtrip times, and against increasing levels of congestion inducing TCP or UDP background traffic on a 10 Gb/s network. Our results show that without much tuning, TCP based protocols are able to achieve throughputs of more than 2 Gb/s. In presence of background traffic, UDP protocols perform better.

Cyberlockers have recently become a very popular means of distributing content. Today, cyberlocker traffic accounts for a non-negligible fraction of the total Internet traffic volume, and is forecasted to grow significantly in the future. The underlying protocol used in cyberlockers is HTTP, and increased usage of these services could drastically alter the characteristics of Web traffic. In light of the evolving nature of Web traffic, updated traffic models are required to capture this change. Despite their popularity, there has been limited work on understanding the characteristics of traffic flows originating from cyberlockers. Using a year-long trace collected from a large campus network, we present a comprehensive characterization study of cyberlocker traffic at the transport layer. We use a combination of flow-level and host-level characteristics to provide insights into the behavior of cyberlockers and their impact on networks. We also develop statistical models that capture the salient features of cyberlocker traffic. Studying the transport-layer interaction is important for analyzing reliability, congestion, flow control, and impact on other layers as well as Internet hosts. Our results can be used in developing improved traffic simulation models that can aid in capacity planning and network traffic management.

We present a comprehensive longitudinal characterization study of the dynamics of content sharing in the global file hosting landscape. We leverage datasets collected from multiple vantage points that allow us to understand how usage of these services evolve over time and how traffic is directed into and out of these sites. We analyze the characteristics of hosted content in the public domain, and investigate the dissemination mechanisms of links. To the best of our knowledge, this is the largest detailed characterization study of the file hosting landscape from a global viewpoint.

There is a continuous struggle for control of resources at every organization that is connected to the Internet. The local organization wishes to use its resources to achieve strategic goals. Some external entities seek direct control of these resources, for purposes such as spamming or launching denial-of-service attacks. Other external entities seek indirect control of assets (e.g., users, finances), but provide services in exchange for them. Using a year-long trace from an edge network, we examine what various external organizations know about one organization. We compare the types of information exposed by or to external organizations using either active ( reconnaissance ) or passive ( surveillance ) techniques. We also explore the direct and indirect control external entities have on local IT resources.

Peer-to-Peer (P2P) applications continue to grow in popularity, and have reportedly overtaken Web applications as the single largest contributor to Internet traffic. Using traces collected from a large edge network, we conduct an extensive analysis of P2P traffic, compare P2P traffic with Web traffic, and discuss the implications of increased P2P traffic. In addition to studying the aggregate P2P traffic, we also analyze and compare the two main constituents of P2P traffic in our data, namely BitTorrent and Gnutella. The results presented in the paper may be used for generating synthetic workloads, gaining insights into the functioning of P2P applications, and developing network management strategies. For example, our results suggest that new models are necessary for Internet traffic. As a first step, we present flow-level distributional models for Web and P2P traffic that may be used in network simulation and emulation experiments.

Wireless local area networks (WLANs) have become omnipresent: WLANs are available at airports, coffee shops, university campuses, corporate environments, and homes. This surge in the popularity of WLANs motivates the study of how these networks are used. Characterizing WLANs, however, is complicated by a number of factors including the geographic diversity of WLAN deployments and the need for capturing activity in the wireless environment instead of the wired environment. In this paper, we describe our experiences with the deployment and use of a remote passive wireless-side measurement infrastructure for monitoring usage of WLANs, and compare our results with a commonly used wired-side measurement technique.

Analyzing traces of wireless network activity has many pragmatic purposes, from capacity planning to network design. Unfortunately, capturing complete traces of wireless traffic is difficult, and using incomplete traces can degrade the quality of the aforementioned analyses. In this paper we examine three different methods for estimating the completeness of wireless traces. We find that a method that examines MAC-layer sequence numbers provides the most accurate results. We also examine the effect of the placement of wireless sensors on the completeness of wireless-side traces. We determine that locating sensors such that the signal strengths between clients and access points is over 40% results in low miss rates at the sensor, and few CRC errors.

This paper describes an interactive graphical user interface tool called Visual Apriori that can be used to study two famous frequent itemset generation algorithms, namely, Apriori and Eclat. Understanding the functional behavior of these two algorithms is critical for students taking a data mining course; and Visual Apriori provides a hands-on environment for doing so. Visual Apriori relies on active participation from the user, where one inputs a transactional database and the tool produces a tree-based frequent itemset generation animation for the algorithm chosen. Visual Apriori provides an effortless learning experience by featuring user-friendly and easy to understand controls.