Sogeti is the Technology and Engineering Services Division of Capgemini. The Sogeti Group is an information technology consulting company specializing in local professional services, with headquarters in Paris, employing around 25,000 people at around 300 branches in 15 countries. The current CEO is Stefan Ek, and in 2011, the company turnover was 1.5 billion euros.
Remote sensing of marine animals is of paramount importance in the context of marine conservation and ecology, as it enables the collection of significant amounts of data automatically. However, manual analysis of such data by experts is highly time consuming. Deep learning models, in contrast to humans, can process vast volumes of visual data in significantly less time and can overcome the variability in detection bias often present in human observers. Some models, such as Convolutional Neural Networks (CNNs), have already surpassed human capabilities in certain visual tasks, including image classification and object detection. However, selecting the appropriate models, techniques, and preprocessing methods that align with the available data type can be a complex task for non-experts, and such choices can significantly impact the quality and reliability of the results. In this study, we provide a systematic review of the literature on various deep learning methods and techniques for the detection of marine animals, discussing the advantages and limitations of each to help researchers select the most appropriate approach for their specific tasks. We also present a compilation of available datasets for marine animal detection and monitoring, while addressing the associated challenges and proposing potential solutions. Finally, we explore emerging research directions and promising approaches aimed at improving the detection of marine animals. In this review, marine animals are defined as marine vertebrates and invertebrates that can be detected or analyzed in satellite, aerial (UAV), and underwater image and video data.
Clean and safe water is a fundamental human right, yet achieving Sustainable Development Goal 6 (SDG 6) ensuring water and sanitation for all remains a persistent global challenge due to pollution, climate stress, limited resources, and aging infrastructure. This study presents a novel, end-to-end AI-quantum hybrid framework that advances water purification, pollution control, and infrastructure monitoring beyond the scope of existing methods. Unlike previous works, which typically focus on isolated tasks or singular model types, we propose the first unified benchmarking pipeline that systematically compares classical, deep learning, and quantum-enhanced models across ten real-world smart water management tasks, including pollutant forecasting, leak detection, microbial risk classification, and anomaly detection. Leveraging large-scale environmental datasets, our models predict contamination trends, optimize treatment protocols, and enable real-time health monitoring of water systems. Quantum models such as Quantum Graph Neural Networks (QGNN), variational quantum circuits, and hybrid CNNs capture high-dimensional, nonlinear relationships that classical models often fail to learn. To enhance transparency and policy integration, we incorporate visualization-driven interpretability tools and ethics-aware deployment strategies. Case studies in arsenic mitigation, heavy metal detection, and microbial purification demonstrate the framework’s real-world applicability and scalability. Overall, this work offers a first-of-its-kind, modular, and ethically aligned AI-quantum architecture designed for resilient and adaptive smart water systems, accelerating measurable progress toward SDG 6, particularly in underserved and resource-constrained regions.
The integration of cloud computing and artificial intelligence (AI) requires advanced data fabric architectures to manage heterogeneous datasets and enable privacy-preserving analytics across multiple domains. This paper introduces a data fabric framework creation proposal for a generic Cloud- AI solution which has multi-domain application as banking, healthcare and utilities through the employment of Variational Autoencoders (VAEs) with differential privacy for synthetic data generation. A comprehensive and well-structured preprocessing pipeline that deals with cross domain data misalignment, achieves 94 (<0.02) supported by illustrative data such as precision recall curves, distribution of data points, privacy-utility trade-off, and aiming for the value of KL-divergence to convince visualizations. The framework achieved 81.58
As ransomware attacks grow in frequency and complexity, accurate attribution is crucial. Victim organizations often feel compelled to pay ransom, but must first attribute the attack and conduct sanction screening to ensure the threat actor receiving the payment is not a sanctioned entity, avoiding severe legal and financial risks. This cyber threat actor attribution process typically relies on Indicators of Compromise (IoCs) matching known threat profiles. However, the emergence of the Ransomware-as-a-Service (RaaS) ecosystem and rebranding behavior complicate attribution for sanction screening. Our mixed-methods study, combining interviews with 20 experts with an analysis of ransomware incident reports, reveals significant challenges and limitations in the current attribution process. High-level IoCs, widely regarded as more reliable, lack the necessary specificity for accurate attribution, leading to potential risks of misattribution. Practitioners rely on lower-level IoCs, which provide clearer links to threat actors but are highly volatile, further complicating sanction enforcement. These challenges highlight the need for urgent improvements in the attribution and sanction processes. To mitigate these risks, we offer recommendations aimed at enhancing data-sharing practices, improving attributions frameworks, and refining the sanction violation policy to better support sanction screening efforts. While we do not recommend paying ransomware actors, we acknowledge that some organizations may face pressures to do so in certain situations. In such cases, it is vital to ensure legal compliance, particularly regarding sanctioned entities. This work aims to help victims of ransomware shield themselves from transgressing against sanctions.
The automotive industry generates vast amounts of data from sensors, telemetry, diagnostics, and real-time operations. Efficient data engineering is critical to handle challenges of latency, scalability, and consistency. Modern data lakehouse formats Delta Parquet, Apache Iceberg, and Apache Hudi offer features such as ACID transactions, schema enforcement, and real-time ingestion, combining the strengths of data lakes and warehouses to support complex use cases. This study presents a comparative analysis of Delta Parquet, Iceberg, and Hudi using real-world time-series automotive telemetry data with fields such as vehicle ID, timestamp, location, and event metrics. The evaluation considers modeling strategies, partitioning, CDC support, query performance, scalability, data consistency, and ecosystem maturity. Key findings show Delta Parquet provides strong ML readiness and governance, Iceberg delivers high performance for batch analytics and cloud-native workloads, while Hudi is optimized for real-time ingestion and incremental processing. Each format exhibits tradeoffs in query efficiency, time-travel, and update semantics. The study offers insights for selecting or combining formats to support fleet management, predictive maintenance, and route optimization. Using structured datasets and realistic queries, the results provide practical guidance for scaling data pipelines and integrating machine learning models in automotive applications.
