IFIP Open Access Digital Library
Not a member yet
22614 research outputs found
Sort by
Unraveling Negation in Modern Greek Using Machine Learning: A Comprehensive Analysis and Detection Framework
No full textPart 2: The 13th Workshop on “Mining Humanistic Data” (MHDW)International audienceNegation detection, a crucial component of natural language processing (NLP), plays a vital role in deciphering sentiment, context, and meaning in textual data. This study delves into the complexity of negation detection in the Greek language, providing a thorough examination of linguistic challenges that set Greek apart from other languages. Greek presents distinctive challenges for negation detection owing to its intricate morphology, complex syntax, and nuanced expressions of negation. In contrast to languages with rigid word order, Greek relies on inflections and particles, demanding NLP models to adeptly navigate diverse linguistic structures. This research reviews prevailing methodologies and tools for negation detection in Greek, encompassing both rule-based and machine learning-based approaches. Through the utilization of linguistic resources and annotated corpora specific to the Greek language, including a X (Twitter) corpus, we assess the performance of these methods, deliberating on their effectiveness in capturing the subtle nuances of negation. It is the first time that negation is being detected in Greek tweets and the annotation process was performed manually. In conclusion, this study offers valuable insights into the inherent challenges of negation detection in the Greek language. It provides a roadmap for researchers and practitioners in the NLP field to craft more precise and context-aware models tailored to the unique linguistic characteristics of Greek
Boosting Data Monetisation with DATAMITE
No full textPart 1: The 9th Workshop on “5G – Putting Intelligence to the Network Edge” (5G-PINE)International audienceCompanies around the globe store large quantities of data they cannot monetise. Regarding internal monetisation, they lack tools to facilitate the governance and quality assessment of their data, resulting not really knowing what data they own or it being non reliable due to its poor quality. External monetisation is usually hindered by the unavailability of trustable mechanisms to perform this exchange or enabling the company to participate in ecosystems like EU data spaces or Gaia-X.DATAMITE is an open-source modular and multi-domain framework that focuses on monetisation through interoperability and data exchange. Its modules offer tools for enhancing data governance, quality and security, but also enabling data sharing to a collection of ecosystems like data spaces, Gaia-X, EOSC or AIoD through a plugin-based approach. It also includes a series of additional support tools to assist on data discovery, ingestion, harmonization or evaluate data fairness, among other
Non-intrusive Weather Analysis for Sustainable Preservation in Cultural Heritage Buildings
No full textPart 3: The 1st Workshop on “AI Applications for Achieving the Green Deal Targets” (ΑΙ4GD)International audiencePreserving the complex structures of our cultural heritage buildings is a primary commitment, as they embody the cornerstone of our history and our common identity. The impact of weather conditions on these buildings is particularly evident. To mitigate the impact of climatic phenomena, innovative solutions utilising advanced technological methods are imperative. This paper proposes a new approach to this complex preservation challenge without compromising historical integrity. This methodology aims to predict wind behaviour and delineate distinct risk exposure zones within cultural buildings and structures using computational fluid dynamics (CFD) models. A case study on the Byzantine-era castle of Mytilene, Greece illustrates the practical application of this approach along with insights crucial for targeted conservation strategies. Concluding, the final section highlights the effectiveness of the proposed methodology, offering prospects for future research and practical applications in heritage conservation
Climate Change and Displacement: Emerging ICT Innovations from the World’s Largest Refugee Camp
No full textPart 6: ICT in Displacement and Conflict Zones: Ideas, Disconnects, and InnovationsInternational audienceThis paper investigates the potential of Information and Communication Technology (ICT) in aiding Rohingya refugees in Cox’s Bazar, Bangladesh, affected by climate change-induced disasters. Focusing on creating tailored solutions, the study highlights challenges faced by refugees and the limited use of ICTs in providing effective support. Collaborating with Young Power in Social Action (YPSA) and their Voice of Palong (VoP) project, the research emphasizes the need for localized early warning systems and communication strategies. VoP’s initiatives, including radio programs and audio-visual content, empower refugees and host communities, enhancing resilience and cohesion. The review examines the concept of climate refugees, ICT’s role in resilience, and vulnerabilities of Rohingya refugees. It stresses the importance of ICTs in monitoring climate change and underscores challenges faced by refugees in vulnerable areas. Discussing emerging themes like communication challenges, mistrust due to misinformation, and ICTs for localized content, the paper concludes by advocating collaborative efforts to utilize ICTs for climate change adaptation and disaster risk reduction. It suggests scaling similar initiatives globally to revolutionize disaster response approaches in various humanitarian settings
The Perspectives of Marginalised South African Youth on a Pre-university Outreach Programme: A Photovoice Study
No full textPart 3: Giving Voice to Marginalised Perspectives in IS ResearchInternational audienceAs part of its mission to reach out to and support students from disadvantaged backgrounds, the University of Pretoria established a programme called the Pre-University Academy (PUA) on its Mamelodi Campus. The PUA hosts courses to complement learners’ school-based learning and further develop their skills so that they can prepare effectively for their tertiary studies. This paper describes how the learners of the PUA utilised the photovoice process, which involves photography, writing and group discussions, to identify the value the programme holds for them. The research used photovoice as a visual data collection method and analysed the qualitative data using reflective thematic analysis. This approach aims to identify patterns of meaning in a dataset and address questions related to peoples’ experiences, views and perceptions. The learners articulated that the PUA held great value for them, both in the present and for their future prospects. They recognised the importance of being part of an academic environment, as the community college setting grants them benefits in terms of both academic and social development
Challenges of ICT4D Solutions in Somalia: A Case Study of Land Management
No full textPart 2: Digital Inclusion Through e-GovernmentInternational audienceThis paper investigates the challenges in Information and Communication Technology for Development (ICT4D) initiatives, wherein they face difficulties when advancing beyond their initial phases, particularly when being deployed across diverse geographical regions. The paper focuses on the interplay among technological integration, local contextual nuances, community engagement, and governmental regulatory frameworks, identifying these as pivotal determinants for ICT4D endeavors. This notion offers value for both scholars and practitioners seeking to enhance the effectiveness of these projects and drive socio-economic advancements within marginalized communities
Assist of AI in a Smart Learning Environment
No full textPart 4: LearningInternational audienceIn recent years and especially during the pandemic period the use of e-learning tools has boomed, moreover, nowadays existing technologies play an important role in providing an excellent online learning experience despite the physical distance between students and teachers and their classmates. An intelligent learning system is a learning environment that uses smart digital devices, new learning approaches, new technologies, and sensors that provide information to artificial intelligence systems. Artificial intelligence makes decisions about adjusting the physical aspects of an intelligent learning environment by analyzing students’ learning performance, behaviors, and the real-life and online situations in which students find themselves. This paper aims to demonstrate the help that artificial intelligence can provide in an intelligent learning system through testing and case studies
Incremental Conflict-Based Search for Multi-agent Path Finding in Dynamic Environment
No full textPart 5: Multi Agent/Ontologies/RoboticsInternational audienceMultiple-Agent Path Finding (MAPF) is a task involving the identification of collision-free paths for a group of agents navigating in a specific map with designated starting vertices and goals. While conventional MAPF focuses on path planning in static environments, real-world scenarios often involve dynamic surroundings, leading to potential blockages in agent paths. In this paper, we propose a multi-agent scheduler Dynamic Incremental Conflict-based Search (DI-CBS) capable of concurrently generating paths satisfying both temporal and spatial constraints in response to environmental changes. A binary Environment Conflict Tree (ECT) is designed for high-level schedule in dynamic environments, effectively addressing determined conflicts and potential conflicts with other agents. Additionally, an incremental search-based planner Spatiotemporal Lifelong Planning A* (SLPA*) is proposed as the low-level module, enabling swift replanning in response to environmental changes. We compare our method with two baseline algorithms on the MAPF benchmark containing 2000 instances. The experiment results verified that our algorithm performs well to a highly dynamic environment
KeepOriginalAugment: Single Image-Based Better Information-Preserving Data Augmentation Approach
No full textPart 1: Deep Learning/ConvolutionalInternational audienceAdvanced image data augmentation techniques play a pivotal role in enhancing the training of models for diverse computer vision tasks. Notably, SalfMix and KeepAugment have emerged as popular strategies, showcasing their efficacy in boosting model performance. However, SalfMix reliance on duplicating salient features poses a risk of overfitting, potentially compromising the model’s generalization capabilities. Conversely, KeepAugment, which selectively preserves salient regions and augments non-salient ones, introduces a domain shift that hinders the exchange of crucial contextual information, impeding overall model understanding. In response to these challenges, we introduce KeepOriginalAugment, a novel data augmentation approach. This method intelligently incorporates the most salient region within the non-salient area, allowing augmentation to be applied to either region. Striking a balance between data diversity and information preservation, KeepOriginalAugment enables models to leverage both diverse salient and non-salient regions, leading to enhanced performance. We explore three strategies for determining the placement of the salient region—minimum, maximum, or random—and investigate swapping perspective strategies to decide which part (salient or non-salient) undergoes augmentation. Our experimental evaluations, conducted on classification datasets such as CIFAR-10, CIFAR-100, and TinyImageNet, demonstrate the superior performance of KeepOriginalAugment compared to existing state-of-the-art techniques. The source code for our KeepOriginalAugment method, along with trained models, is publicly available at https://github.com/kmr2017
Adaptive Image Adversarial Example Detection Based on Class Activation Mapping
No full textPart 5: ML Attack, VulnerabilityInternational audienceWith the development of deep learning technology, convolutional neural network (CNN) has been widely used in many fields such as face recognition, automatic driving, biomedicine, etc., replacing human beings to complete complex and redundant work, which brings great convenience to people’s lives. However, the discovery and development of adversarial examples have created a greater threat to image recognition. In this paper, we propose an adaptive image adversarial example detection method based on class activation mapping, which utilizes the hot zone discovery results of the Grad-CAM algorithm to perform adaptive noise reduction on images and analyzes the differences in the classification results of images before and after the noise reduction in the same benchmark network, including the KL dispersion, the label change, the label confidence, etc., to achieve the detection of adversarial examples on the ImageNet-1000 dataset. The experimental results show that the algorithm proposed in this paper achieves better detection results, and the F1 reaches 0.82 in detecting the generated FGSM adversarial examples with ϵ = 0.3, which is better than the baseline model