Exploring AI Technologies from Ethics to Trust (ENFIELD educational material)

The 2024 ENFIELD webinar was organized in the framework of the Horizon Project ENFIELD, having taken place online on November 15, 2024.

The objective of the webinar was to discuss the critical landscape of AI ethics and trust.

During the webinar, 8 presentations were given:

• Pankaj Pandey – Presentation of the ENFIELD project
• Jean Gabriel Ganascia – Ethical Implications of AI Technologies in Society
• Christof Wolf-Brenner – Societal Impact and Ethical Challenges of AI
• Jeriek Van den Abeele – Responsible AI from Research to Practice
• Asbjørn Følstad – Ethics and trust from a human-centric perspective
• Phillip Morgan – How to build trust in AI
• Ștefan Trăușan-Matu – Ethics and trust in AI Natural Language Processing
• Natalia Glowacka – ENFIELD Ethics and Gender Equality Framework

X-ray Anomaly Detection in Industrial Pipelines

As Deep Neural Network (DNN)-based algorithms are improving, pivotal changes are happening towards efficient and effective automation in the field of industrial inspection. In the scope of our project, we analyze X-ray images of steel pipelines to detect the presence of corrosion in a novel way. In our industrial scenario, a drone lands a crawler that is equipped with an X-ray system on top of insulated pipelines to perform X-ray scans which are able to penetrate only the insulation, due to power consumption limitations. In this paper, we use modern unsupervised anomaly detection algorithms to detect the presence of corrosion, and the results are quite promising. Moreover, to compare several state-of-the-art approaches in terms of robustness to noise, we simulate two types of noise that can occur: (i) Poisson Noise, (ii) Motion Blur Noise. We conclude that the problem we are dealing with can be handled sufficiently well with state-of-the-art approaches, and that in the scenario of noise, the most robust algorithms are based on memory banks and teacher-student architectures.

Foreground-Aware Knowledge Distillation for Enhanced Damage Detection

Damage detection remains a critical challenge, especially within the industrial automation sector, necessitating the development of advanced inspection technologies and their potential applications. Conventional industrial inspection methods are hindered by high costs and operational disruptions, motivating the development of innovative and efficient solutions. This paper introduces a novel, architecture-agnostic deep neural network (DNN) knowledge distillation (KD) method able to enhance vision-based damage detection performance even in challenging industrial environments. Our proposed method integrates foreground knowledge with feature KD to enhance data feature utilization in detection models, effectively minimizing background clutter. The results demonstrate the efficiency of our method in consistently enhancing the student’s training process, including up to a 12% increase in mean Aver-age Precision (mAP), across various DNN architectures. Our approach bridges the gap between academic research and real-world industrial cur-rent applications, offering a robust solution for damage detection in insulated pipelines.

Advancing Industrial inspection: A Dataset for Automated Damage Detection in Insulated Pipes

The application of automated inspection for industrial pipe damage detection is attracting substantial research and development interest. Damage to pipes not only hinders the optimal functioning of factories but also presents a risk of industrial disasters, making the adoption of automated solutions imperative. The use of Unmanned Aerial Vehicles (UAVs) equipped with Deep Neural Network (DNN)-enhanced vision offers an innovative method to detect pipe damage in real-time or during video post-processing. However, a major challenge in fully leveraging the capabilities of DNNs for pipe damage detection is the lack of specialized, well-structured, and annotated public datasets for DNN training. This paper introduces the Pipes Damages Image (PDI) dataset, the first of its kind, specifically designed for detecting damage in insulated industrial pipes. This carefully compiled dataset covers a wide range of industrial settings, with each scenario meticulously annotated for damage detection. It also provides base-line results from state-of-the-art visual object detection models.

Undergraduate University AI Education: A Survey

Artificial Intelligence (AI) has become a pivotal technology of the 21st century, prompting the rapid development of undergraduate and postgraduate AI education programs worldwide. This paper presents a comprehensive survey of these programs, spanning the historical evolution of undergraduate AI education and revealing global trends. Undergraduate AI education equips the future workforce with fundamental AI knowledge to harness its potential across diverse sectors. Therefore, many countries paid due attention to University AI education. Overall, Global North countries fare very well in University AI education. China has emerged as a prominent leader in AI education, driven by its strategic national plan. A comparative analysis of renowned Universities showcases the structure of AI curricula, emphasizing the need to balance theory and application. Overall, this paper is a valuable resource for stakeholders interested in the evolving landscape of University AI education.

Decentralized DNN Architectures

This lecture overviews decentralized and distributed DNN architectures. Big data analysis can be greatly facilitated if decentralized/distributed DNN architectures are employed that interact with each other for DNN training and/or inference using the human Teacher-Student education paradigm. A novel Learning-by-Education Node Community (LENC) framework is presented that facilitates communication and knowledge exchange among diverse Deep Neural Networks (DNN) agents, undertaking the role of a student or teacher DNN by offering or absorbing knowledge respectively. The framework enables efficient and effective knowledge transfer among participating DNN agents while enhancing their learning capabilities and fostering their collaboration among diverse networks. The proposed framework addresses the challenges of handling diverse training data distributions and the limitations of individual DNN agent learning abilities. The LENC framework ensures the exploitation of the best available teacher knowledge upon learning a new task and protects the DNN agents from catastrophic forgetting. The experiments demonstrate the LENC framework functionalities on multiple teacher-student learning techniques and their integration with lifelong learning. Our experiments manifest the LEMA framework’s ability to maximize the accuracy of all participating DNN agents in classification tasks by leveraging the collaborative knowledge of the framework. The LENC framework also addresses the problem of task-agnostic lifelong learning as DNN agents have no information on task boundaries.