Explainable artificial intelligence (XAI) plays a crucial role in mitigating the risks associated with the non-transparency of black-box artificial intelligence (AI) systems. However, despite its advantages, XAI methods have been shown to expose the privacy of individuals whose data are used to train or query the underlying models. Prior research has demonstrated privacy attacks that exploit explanations to infer sensitive personal information of individuals. At present, there is a lack of effective defenses against such privacy attacks targeting explanations, particularly when vulnerable XAI techniques are deployed in production environments or used in machine learning as a service systems. To address this gap, this study investigates the use of privacy enhancing technologies (PETs) as a defense mechanism against attribute inference attacks on explanations generated by feature-based XAI methods. We empirically evaluate three types of PETs, i.e., synthetic training data, differentially private training and noise addition, across two categories of feature-based XAI. Our findings reveal varying levels of effectiveness among the mitigation strategies, as well as trade-offs between privacy, utility and system performance. In the best scenario, integrating PETs into the explanation process reduced attack success by 49.47% while preserving model utility and explanation quality. Based on our evaluation, we propose strategies for effectively integrating PETs into XAI to maximize privacy protection and minimize the risk of sensitive information leakage.

Backyard poultry operations pose a potential risk of avian influenza transmission, a disease with severe economic consequences due to flock culling and trade restrictions. Existing surveillance efforts rely primarily on data from registered farms, often overlooking unregistered small-scale flocks that lack biosecurity and are more exposed to wild birds, a known reservoir for the virus. This creates a gap in the monitoring of avian influenza. This study proposes a deep learning-based approach specifically designed to detect backyard operations using high resolution satellite imagery. Although previous studies have applied satellite imagery and deep learning techniques to detect commercial poultry farms and large-scale livestock operations, these approaches have not been extended to backyard poultry detection. Our method addresses the challenge of identifying small, irregular and often unregistered backyard setups. We employ a fully convolutional network (FCN) with ResNet-50 backbone to perform binary semantic segmentation. The model achieved an accuracy of 81.13%, precision of 78.92%, recall of 84.96%, and an F1 score of 81.83%, outperforming other models on our dataset.

Black-box Artificial Intelligence (AI) systems have achieved state-of-the-art accuracy in many problem domains in recent years. However, the lack of transparency of these systems is a bottleneck in their usage in high-risk applications which make automated decisions on individuals. Trustworthy AI proposes principles such as reliability, validity, privacy, fairness, and explainability among others, to mitigate risks from large-scale AI deployments in such domains. Explainable AI (XAI) is a technique of providing insights into the decision-making process of black-box systems thus enabling transparency. It plays a crucial role in communicating the rationale of automated decisions to relevant stakeholders. Though explainability is a highly desirable requirement, recent research has determined that explanations can introduce new privacy risks in AI systems. Researchers have demonstrated different types of privacy attacks on XAI deployed in production and cloud systems. Despite these risks, currently there is a lack of research into defenses for known privacy attacks in XAI. In this article, we contribute to this gap by proposing a defense mechanism for attribute inference attack on feature-based XAI. We empirically evaluate a well-known privacy preservation technique, namely, additive noise, and show its impact on privacy, explainability and utility. Our findings indicate that additive noise enables privacy while achieving faithful explanations and without compromising model utility.

Privacy attacks on machine learning (ML) models pose significant risks to individuals whose personal data is used for training or querying these models. Although concerns about the potential exposure of sensitive information through ML models continue to grow, existing safeguard mechanisms primarily focus on security threats, often neglecting privacy risks. In this paper, we examine existing tools to assess privacy risks of ML models and provide an overview of various privacy attacks and defense strategies. Given the lack of a comprehensive framework for assessing privacy vulnerabilities, we propose a generic framework for evaluating the privacy of ML systems and establish a set of tailored evaluation metrics for different types of privacy attacks. In addition, we develop a dedicated testbed to implement our framework and present experimental results that demonstrate the impact of various privacy attacks on different ML models.

The emergence of smartphones and internet accessibility around the globe have enabled billions of people to be connected to the digital world. Due to the popularity of instant messaging applications and social media, a large quantity of personal data is in text format, and processing text data in a privacy-preserving manner poses unique challenges. While existing reviews focus on privacy concerns from specific algorithmic perspectives or target only a particular domain, such as healthcare or smart metering, they fail to provide a comprehensive view that addresses the multi-layered privacy risks inherent to text data processing. Existing works often limit their scope to specialized solutions like differential privacy, anonymization, or federated learning, neglecting a broader spectrum of challenges. To fill this gap, we present a comprehensive review of privacy-enhancing solutions for text data processing in the present literature and classify the works into six categories of privacy risks: (i) unintentional memorability, (ii) membership inference, (iii) exposure and re-identification, (iv) language models and word embeddings, (v) authorship attribution, and (vi) collaborative processing. We then analyze existing privacy-enhancing solutions for text data by considering the aforementioned privacy risks. Finally, we identified several research gaps, including the need for comprehensive privacy metrics, explainable algorithms, and privacy in social media analytics.

Avian Influenza Virus (AIV) poses significant threats to the poultry industry, humans, domestic animals, and wildlife health worldwide. Monitoring this infectious disease is important for rapid and effective response to potential outbreaks. Conventional avian influenza surveillance systems have exhibited limitations in providing timely alerts for potential outbreaks. This study aimed to examine the idea of using online activity on social media, and Google searches to improve the identification of AIV in the early stage of an outbreak in a region. To this end, to evaluate the feasibility of this approach, we collected historical data on online user activities from X (formerly known as Twitter) and Google Trends and assessed the statistical correlation of activities in a region with the AIV outbreak officially reported case numbers. In order to mitigate the effect of the noisy content on the outbreak identification process, large language models were utilized to filter out the relevant online activity on X that could be indicative of an outbreak. Additionally, we conducted trend analysis on the selected internet-based data sources in terms of their timeliness and statistical significance in identifying AIV outbreaks. Moreover, we performed an ablation study using autoregsressive forecasting models to identify the contribution of X and Google Trends in predicting AIV outbreaks. The experimental findings illustrate that online activity on social media and search engine trends can detect avian influenza outbreaks, providing alerts earlier compared to official reports. This study suggests that real-time analysis of social media outlets and Google search trends can be used in avian influenza outbreak early warning systems, supporting epidemiologists and animal health professionals in informed decision-making.

In recent times, machine learning and deep learning algorithms have contributed to the advances in audio and speech recognition. Despite the progress, there is limited emphasis on the classification of cantillation audio using deep learning. This paper introduces a dataset containing two labeled styles of cantillation from six reciters. Deep learning architectures including convolutional neural networks (CNN) and deep artificial neural networks (ANN) were used to classify the recitation styles using various spectrogram features. Moreover, the classification of the six reciters was also performed using deep learning. The best performance was achieved using a CNN model and Mel spectrograms resulting in an F1-score of 0.99 on the test set for classifying recitation style and an F1-score of 1.00 on the test set for classifying reciters. The results obtained in this work outperform the existing works in the literature. The paper also discusses the impact of various audio features and deep learning algorithms that apply to audio genre and speaker identification tasks.

Privacy policies are statements about how websites, applications, and any other service providers collect, use, share and manage users' data. Nowadays, the contents of privacy policies have been affected by different regulations such as the General Data Protection Regulation (GDPR), which enforces the protection of personal data and also requires privacy policies to be more transparent for readers. There is a limited understanding of how GDPR has impacted the content of privacy policies. This study presents a framework for evaluation of compliance of privacy policies with GDPR recommendations and best practices. This evaluation framework includes text feature analysis, coverage analysis, and content analysis. Our findings suggest that although GDPR enforcement has improved the content of privacy policies, many of these privacy policies do not fully satisfy GDPR requirements.

Privacy policy is a legal document in which the users are informed about the data practices used by the organizations. Past research indicates that the privacy policies are long and hard to understand. They are also known to have incomplete content. Users are not inclined to read the policy as they have to read long policies to find information about data practices of an organization. The solution that we are proposing in this research is to assist users with finding relevant content to their queries using semantic approach. This thesis presents the development of domain ontology for privacy policies. Natural Language Processing was used to understand the content of the policies and capture vocabulary for the ontology. This vocabulary was further used to build the ontology so that the ontology highlights relevant sentences related to a privacy concern. We validated and evaluated the ontology using different methods: competency questions, data driven, metric based and user evaluation. Results from the evaluation of ontology show that the amount of text to read is significantly reduced as the users have to only read selected text that ranged from 1% to 30% of a privacy policy. The amount of text depended on the query and its associated keywords. This signifies that the time required to read a policy is significantly reduced as the ontology directs user to the right content for a query. This finding was also confirmed by the results of the user study session. The results from the user study session indicated that the users found ontology helpful in finding relevant selected sentences to read as compared to reading the entire policy.

The expectation on the agricultural system is constantly growing to be more productive with less labor, less water and less arable land. To achieve this goal, the use of digital technologies is being promoted. This has resulted in growth in use of wireless sensors, IoTs, cloud computing and other technologies in farms which have fueled the explorational of data. Data collected at farms varies from business operation data (farm management data), transport and farm storage data, land data (water, soil, GPS), machine data, agronomic data, livestock data, climate, and weather data. This large amount of data needs to be managed to ensure confidentiality and other governance requirements and enhance technical capacity and performance such as data integration and processing. This paper reviews the data governance challenges generated in smart farms and provides recommendations on how those challenges can be addressed.

Personal information protection and compliance with privacy regulations are becoming increasingly important due to a large number of security and privacy breaches. These privacy breaches can harm individuals in both personal and social contexts. Privacy policies are the primary means of communication with which service providers could inform users about the data collection and sharing practices. The content and transparency of these legal documents are of importance as they can help users make decisions about the service providers’ data privacy practices and can build trust with the users. Although many regulations and best practices have provided recommendations and guidelines on the content of privacy policies, research has shown that the content of these documents is usually incomplete and miss important topics. To address this issue, we propose and validate the use of automated generative models for creating the content of privacy policies. These generative approaches use deep learning models to generate enriched data practices and text for privacy policies. In this study, we trained two generative models (Long Short-Term Memory (LSTM) and bidirectional Long Short-Term Memory bi-LSTM) on annotated privacy policies to automatically generate privacy data practices. Training and testing were performed on three levels of paragraph, sentence, and data practice. Our findings have shown promising results and have suggested that models trained on legal data practices using bi-LSTM algorithm create more accurate results.

This session explores how Complex Adaptive Systems provide a framework for analyzing important social, biological, and environmental systems in One Health. Anthropogenic disturbances, many of which are technological, pose a threat to key ecological and sociological processes. They lead us to consider questions such as: Is artificial intelligence a saviour or a demon? What are the political, ethical, and scientific implications for One Health? How might the Global Burden of Disease (human), the Global Burden of Animal Diseases (GBADs) and other Global Burdens constitute a broader “One Health Burdens of Disease” and provide an evidence-base for One Health decisions? It will be necessary to address different data challenges in the developed and developing worlds, many of which are ethical and political, not just technical. Panelists will discuss the GBADs approach to data sharing, including how FAIR-principled metadata can be used to create trustworthy data systems and how the Data Governance Handbook provides important guidance for communicating data sharing principles to data contributors and users. Each panelist will provide a 5-10 minute “primer” talk which will introduce and link the key themes. This will be followed by a moderated panel discussion with opportunities for the audience to pose questions.

Personal Identification Numbers (PINs) are widely used as an access control mechanism for digital assets (e.g., smartphones), financial assets (e.g., ATM cards), and physical assets (e.g., locks for garage doors or homes). Using semi-structured interviews (n=35), participants reported on PIN usage for different types of assets, including how users choose, share, inherit, and reuse PINs, as well as behaviour following the compromise of a PIN. We find that memorability is the most important criterion when choosing a PIN, more so than security or concerns of reuse. Updating or changing a PIN is very uncommon, even when a PIN is compromised. Participants reported sharing PINs for one type of asset with acquaintances but inadvertently reused them for other assets, thereby subjecting themselves to potential risks. Participants also reported using PINs originally set by previous homeowners for physical devices (e.g., alarm or keypad door entry systems). While aware of the risks of not updating PINs, this did not always deter participants from using inherited PINs, as they were often missing instructions on how to update them. Given the expected increase in PIN-protected assets (e.g., loyalty cards, smart locks, and web apps), we provide suggestions and future research directions to better support users with multiple digital and non-digital assets and more secure human-device interaction when utilizing PINs.

A number of methods have been proposed for the construction of vector representations for natural language processing (NLP) tasks. These methods have been applied to various domains and each has its own pros and cons. Despite their effectiveness, the proposed approaches usually ignore the sentiment information concerning specific tasks. In this paper, we examined various types of word vectors and their impact on the performance of a sentiment classification problem in the area of infectious diseases. Vectors were used in the embedding layer of a word-based convolutional neural network (CNN) to identify tweets pertaining to avian influenza. We proposed a new approach to build effective word embeddings for the sentiment analysis task. Furthermore, the performance of the language model was compared in terms of using various corpus sizes and vector dimensions. Our experiments indicated that initializing the sentiment learning network with domain-specific word embeddings outperforms general domain embeddings. We found that the proposed method leads to a considerable improvement in the classification performance.

Epidemic Avian Influenza (AI) outbreaks pose a considerable threat to poultry and human health. Millions of birds died, or had to be euthanized, during numerous periodic highly pathogenic influenza outbreaks across the globe over the last two decades. This paper presents a study on pathogenicity classification of H5 AI protein sequences using convolutional neural network. We first collected pathogenicity identification labels of 2137 H5 AI sequences and classified pathogenicity of aligned protein sequences with 99.20% mean accuracy using 10-fold cross validation. We also identified the positions in genomic sequences which are supposed to play a role in pathogenicity classification of H5 strains. The positions learned by convolutional network correctly included positions of known pathogenicity markers present at cleavage site of protein sequences. And the network also identified a few other positions present outside of cleavage site which are supposed to play a role in pathogenicity classification of H5 strains.

Keywords and keyphrases are important for Natural Language Processing (NLP) applications such as document classification, information retrieval, and topic identification. They are also useful for capturing different classes of entities from content related to healthcare, biology, food science, and journalism fields. There are different approaches to extract keywords and keyphrases. Deep learning approaches have achieved high-performance results in terms of keywords and keyphrase extraction. However, among deep learning approaches, Convolutional Neural Network (CNN) potentials have not been fully explored as a technique for extracting keywords and keyphrases. In this work, we performed a comparative study using a benchmark dataset, the IEEE Xplore collection to test the CNN generalization ability in selecting keywords and keyphrases. In addition, we further collected a corpus in the field of foodborne illness outbreaks. We utilize this corpus to develop a CNN-based identification approach of keywords and keyphrases related to foodborne illnesses. Results were compared with several supervised (KEA, GuidedLDA) and unsupervised (LDA) machine learning algorithms. CNN outperformed these algorithms in selecting relevant keywords and keyphrases for foodborne illnesses. The findings of this study have also confirmed superiority of CNN-based algorithm for keyphrase extraction to other machine learning approaches.

Web site privacy policies are often long, difficult to understand, and contain incomplete information. Consequently, users tend not to read the privacy policies, thus putting their privacy at risk. This paper describes an automated approach for assisting users to evaluate online privacy policies based on completeness. The term completeness refers to the presence of 8 sections in an online privacy policy that have been recognized as helpful in establishing the transparency of a privacy policy. Given a new online privacy policy, the proposed system employs a machine-learning based approach to predict a completeness score for the privacy policy. This score can then be used by the user to assess the risk to their privacy.

Feature selection is still a vital area for research in the machine learning field. After the emergence of big data, the need for mining large data sizes has increased to provide faster and more accurate predictions. Feature selection is concerned with selecting the most important features from a set of input features since some datasets may contain irrelevant and/or redundant features. In this paper, a new feature selection method of type embedded is presented and discussed with some preliminary results using existing benchmark datasets. The new method is called Recursive Feature Addition which works in a forward fashion and is based on Support Vector Machines. The new method has been applied to five different benchmark datasets and for which it has shown superior performance in terms of accuracy and time as compared to Filter, Wrapper and other Embedded methods.

Data partitioning methods such as bagging and boosting have been extensively used in multiple classifier systems. These methods have shown a great potential for improving classification accuracy. This study is concerned with the analysis of training data distribution and its impact on the performance of multiple classifier systems. In this study, several feature-based and class-based measures are proposed. These measures can be used to estimate statistical characteristics of the training partitions. To assess the effectiveness of different types of training partitions, we generated a large number of disjoint training partitions with distinctive distributions. Then, we empirically assessed these training partitions and their impact on the performance of the system by utilizing the proposed feature-based and class-based measures. We applied the findings of this analysis and developed a new partitioning method called "Clustering, Declustering, and Selection" (CDS). This study presents a comparative analysis of several existing data partitioning methods including our proposed CDS approach.

The testing activities of the Software Verification and Validation (SV&V) team at Research In Motion (RIM) are requirements-based, which is commonly known as requirements-based testing (RBT). This paper proposes a novel approach to enhance the current RBT process at RIM, by utilizing historical testing data from previous releases, static analysis of the modified source code, and real-time execution data. The main focus is on the test case generation phase and the objective is to increase the effectiveness and efficiency of test cases in such a way that overall testing is improved. The enhanced process not only automatically generates effective test cases but also seeks to achieve high test coverage and low defect escape rate.

The need for the design of complex and incremental training algorithms in multiple neural network systems has motivated us to study combining methods from the cooperation perspective. One way of achieving effective cooperation is through sharing resources such as information and components. The degree and method by which multiple classifier systems share training resources can be a measure of cooperation. Despite the growing number of interests in data modification techniques, such as bagging and k-fold cross-validation, there is no guidance for whether sharing or not sharing training patterns results in higher accuracy and under what conditions. We implemented several partitioning techniques and examined the effect of sharing training patterns by varying the size of overlap between 0-100% of the size of training subsets. Under most conditions studied, multinet systems showed improvement over the presence of larger overlap subsets.