IFIP Open Access Digital Library
Not a member yet
22614 research outputs found
Sort by
Advancing Agricultural Sustainability Through an AI Powered Classification Framework of Plant Pests and Diseases
No full textPart 3: The 1st Workshop on “AI Applications for Achieving the Green Deal Targets” (ΑΙ4GD)International audiencePlants pests and diseases can cause significant economic losses for farmers, and accurate identification and classification of these issues is crucial. Deep learning techniques, have demonstrated promising results in correctly identifying and classifying plant pests and diseases. However, the accuracy of these models heavily relies on the quality and size of the training dataset. Gathering a large and diverse dataset of plant pests and diseases can be challenging due to the variation in the appearance of diseased plants and the high cost of data collection. To tackle this issue, a web portal has been developed to gather datasets of plant pests and diseases, making it simpler for researchers to contribute data and enhance the accuracy of these models. In addition, Generative Adversarial Networks (GANs) have been used to expand the current datasets by creating synthetic images that closely resemble real-world data. This has resulted in improved accuracy and robustness of CNN and YOLO models in identifying and classifying plant pests and diseases
Vertical Federated Image Segmentation
No full textPart 1: Deep Learning/ConvolutionalInternational audienceThere has been a growing concern for both data privacy and acquisition for the purpose of training robust computer vision algorithms. Often, information is located on separate data silos and it can be difficult for a machine learning engineer to consolidate all of it in a fashion that is appropriate for model development. Additionally, some of these localized data regions may not have access to a labelled ground truth, rendering conventional model training impossible. In this paper, we propose a novel architecture that can perform image segmentation in vertical federated environments. This is the first implementation of a federated architecture that can perform vertical federated image segmentation. We utilized a distributed vertical fully convolutional network architecture that is able to train on data where the segmentation maps reside on a different federate than the original image. Our architecture is able to compress the features of an image from 49,152 down to 500, allowing for more efficient communication between the top and bottom model of our federated architecture. We trained our model on 369 images from the CamVid dataset. Our model demonstrates a robust capability for accurate road detection
A Multi-scale Parallel Unsupervised Model for Multivariate Time Series Anomaly Detection
No full textPart 4: LearningInternational audienceAnomaly detection for multivariate time series data is of great significance for practical applications. The existing anomaly detection methods mainly adopt a fixed length sliding window to extract data features and perform deep learning training. However, a single fixed length window of data makes it difficult to simultaneously detect anomalies in different scale, such as small-scale point anomalies and large-scale contextual anomalies. Additionally, the patterns in multivariate time series data may be more complex and diverse. This paper proposes an unsupervised multi-scale model for multivariate time series anomaly detection (MMTSAD) to address the above problem. We downsample the original data to obtain coarse-grained and fine-grained sequences in different scales, and design two autoencoder modules based on attention mechanism to learn time series patterns at different scales. In addition, we introduce a GAT module in the coarse-grained autoencoder to capture the correlation between different variables. At the detection stage, we propose an anomaly score fusion method to comprehensively fuse the anomaly scores from different scale models. We conduct experiments on five real-world public datasets. The results show that MMTSAD outperforms most existing models
Modeling the Air Conditioner Performance Tests Using Artificial Neural Network Simulator (ANNS-AC)
No full textPart 2: Recommendation/ClassificationInternational audienceThe aim of the present study was to use the artificial neural network (ANN) to simulate the performance of air condition (AC) unit for validation purposes. This helps save time and effort instead of repeating the test for validation. A backpropagation ANN models with multiple hidden layers were trained using 22 input variables and three targets. More than 800 test reports were used to train the ANN model. The input processing functions, neuron sizes, starting values of the weights and biases, layer transfer functions, training functions, and performance evaluation functions were discussed. The uncertainty components associated with the experimental measurements and learning leakage in the ANN model were evaluated. It was found that the ANN model can predict the performance of AC unit under certain experimental conditions using the information provided by the manufacture. However, to achieve a reliable result, the output of the ANN model should be evaluated as an average of at least 50 runs, with reinitiating the starting values of the weights and biases in each run. The model was also used to study the effect of airflow on the performance of the AC and identify the conditions leading to high AC efficiency. The results indicated that under specific conditions, the AC can achieve maximum efficiency without increasing the power input
Active Learning with Unfiltered Informativeness Technique for Object Detection
No full textPart 4: LearningInternational audienceContemporary Deep Learning models demand substantialvolumes of data to effectively learn, creating a challenge given the difficultyof obtaining well-annotated data. Moreover, not all samples withina dataset are of equal significance to the learning process. Active Learningemerges as a solution to this issue by providing a structured frameworkfor selecting the most instructive data within a dataset. This approachinvolves isolating a subset of the N most informative samples totrain the algorithm effectively. In response to this challenge, we introduceUnfiltered Informativeness, a novel framework designed to assess theinformativeness value of samples within a dataset. Our approach employsa trained detector to identify objects in a scene and subsequently computesthe information gain of these objects. When multiple objects arepresent within a scene, we aggregate multiple scores into a single informativescore. We systematically evaluate our approach against RandomSampling and other strategies
A Graph-Based Framework for ABAC Policy Enforcement and Analysis
No full textPart 1: Access ControlInternational audienceIn the realm of access control mechanisms, Attribute-Based Access Control (ABAC) stands out for its dynamic and fine-grained approach, enabling permissions to be allocated based on attributes of subjects, objects, and the environment. This paper introduces a graph model for ABAC, named GABAC. The GABAC leverages directional flow capacities to enforce access control policies, mapping the potential pathways between a subject and an object to ascertain access rights. Furthermore, graph based modeling of ABAC enables the utilization of readily available commercial graph database systems to implement ABAC. As a result, enforcement and analyses of ABAC can be accomplished simply through graph queries. In particular, we demonstrate this using the Neo4j graph database and present the performance of executing enforcement and different analyses queries
Visor: Privacy-Preserving Reputation for Decentralized Marketplaces
No full textPart 3: PrivacyInternational audienceFeedback mechanisms are crucial in e-commerce and collaborative systems, shaping trust, deterring dishonesty, and promoting good online behavior. Reputation, built through collective feedback, impacts vendor trustworthiness and support. However, centralized systems like eBay and Amazon pose privacy concerns as users must trust the central server to maintain accurate records and protect sensitive data. Decentralized solutions offer greater privacy and control over feedback, aiming to overcome the limitations of centralized ones. Yet, decentralized reputation systems face challenges such as susceptibility to Sybil attacks and more, which undermine user trust and system reliability.In this work, we propose Visor, a decentralized reputation system addressing these challenges by leveraging randomizable signatures and zero-knowledge proofs to construct a reputation mechanism that safeguards privacy and prevents misuse. The security properties of Visor are formally demonstrated; the system guarantees integrity and ensures that users remain anonymous during feedback, while also maintaining unlinkability among pseudonyms and reviews associated with the same user. Finally, the system provides users with a comprehensive view of all transactions, thereby enabling public verifiability without reliance on any trusted third party
Exploring Data Altruism as Data Donation: A Review of Concepts, Actors and Objectives
No full textInternational audienceThis paper reviews data altruism in the emerging academic literature by collecting and analyzing conceptually similar terms, identifying key actors, and delineating the objectives of general interest underlying this novel form of voluntary data sharing. Drawing from a wide array of disciplines including computer science, social science, law, and medicine and bioethics, we discover frequent comparisons with the notions of data donation, data crowdsourcing, data philanthropy, and data solidarity. In our observations of current definitions and understandings of data altruism and these related terms, we draw attention to analyzing their salient similarities and differences. Our analysis of actors and roles illustrates a wide variety of players envisioned to participate in data altruism. Our examination of different types of data discussed in the literature suggests the data altruism value chain proposition is largely promissory, with limited empirical evidence supporting specified objectives of general interest. This review contributes to a refined understanding of data altruism as a formal ‘data donation’ institutionalization process in the European digital space and highlights its implications for future research
Ethical Governance of Emerging Digital Technologies in the Public Sector
No full textInternational audienceEmerging digital technologies, such as algorithms and machine learning, offer transformative opportunities to public sector organizations but also pose ethical risks and dilemmas. Public sector organizations adopting these technologies need to govern their ethical implications. This research provides insights into the emerging phenomena of digital ethics commissions within Dutch public sector organizations. Composed of external experts on ethics, technology, and governance, these commissions are meant to reflect, advise, and, in some cases, assess the ethical design and use of emerging digital technology and the governance thereof. Through interviews and document analysis, this research explores the motivations, intentions, and perceptions guiding these commissions. Our preliminary findings suggest that, while digital ethics commissions are meant to convey legitimacy, they can also provide external knowledge, open the organizations for reflection from and with society, and contribute to different types of control. In establishing these commissions, government organizations need to balance the formalization of digital ethics governance with the need for a collaborative and reflective ethical practice conducive to (organizational) learning and ethics as contextual practice. This research contributes empirical insights into how public sector organizations address ethical challenges from emerging digital technologies, offering valuable implications for both practitioners and scholars in the field of public administration
A Method for the Collaborative and Semi-automated Generation of Conceptual Models from Legal Regulations in Public Organizations
No full textInternational audienceLegal regulations and conceptual models are important for public organizations. Conceptual models are means for complexity reduction in the design and customization of information systems. Legal regulations are important for public organizations as they specify the services that the organizations offer to citizens and businesses. However, the operationalization of legal regulations is challenging because they leave room for interpretation and there is a high number of involved actors. These actors comprise of domain experts and IT experts within a public organization, but also other public organizations on the same or different levels of government. In consequence, there are many actors involved in the operationalization and execution of legal regulations for public services who would benefit from the use of conceptual models. To provide support for these actors, we address the following research goal: Design of a method for the collaborative and semi-automated generation of conceptual model from legal regulations in public organizations. In the course of our design science research approach, we derived requirements for the solution based on interviews with thirteen public officials. We conceptually developed the method and evaluated it with an illustrative scenario and expert interviews with six public officials. The evaluations reveal the general potential usefulness and intended use of our method