Abstract
This study analyzes forest cover change patterns, agricultural expansion, and economic growth in Mozambique from 2001 to 2024, using remote sensing data from Global Forest Watch and socioeconomic indicators from the World Bank and FAO. Mozambique lost approximately 4.6 million hectares of forest during the analyzed period, with agriculture accounting for 97.4% of total deforestation. GDP per capita increased by 90.5%, while cultivated area expanded by 116.4%. However, agricultural productivity declined by 25.3%, revealing a paradox: production growth relied on extensive land expansion rather than intensification. Statistical analysis of three 8-year sub-periods identified significant differences in GDPpercapita, agricultural GDP per capita, population, and agricultural employment (p < 0.001), but agricultural deforestation remained statistically stable (p = 0.065). This pattern suggests premature decoupling between economic growth and deforestation at income levels (USD 604) substantially below historical Environmental Kuznets Curve thresholds (USD 8000–10,000). However, this decoupling is fragile, driven by capital-intensive extractive sectors that generate GDP growth without absorbing rural populations. The persistence of extensive agricultural expansion, combined with weak governance, demographic pressures, and climate variability, indicates that observed stabilization represents an initial, vulnerable phase requiring structural transformation through agricultural intensification, inclusive industrialization, land tenure reform, and climate resilience building.

Abstract
Machine learning (ML) frameworks, such as PyTorch and TensorFlow, rely on data loaders to preprocess data before feeding it to accelerators. When preprocessing is inefficiently pipelined, GPUs can remain idle over long periods of time, leading to substantial training delays. For example, PyTorch's default data loaders can cause up to 76% GPU idleness. A key bottleneck is the variability in preprocessing time across samples within the same dataset. Existing data loaders are oblivious to this variability, training all samples uniformly. In this case, a single slow sample can stall the entire batch, causing head-of-line blocking. We present MINATOLOADER, a general-purpose data loader for PyTorch that accelerates training and improves GPU utilization under single-server, multi-GPU settings. It continuously prepares data in background and constructs batches by prioritizing fast-to-process samples, while slower samples are processed in parallel. Experiments conducted overNVIDIA V100 and A100 GPUs show that MINATOLOADER accelerates training by up to 7.5x (3.6x on average) over PyTorch DataLoader and Pecan, and up to 3x (2.2x on average) over DALI. It also increases average GPU utilization from 46% with PyTorch to 90%, while preserving model accuracy and enabling faster convergence.

Abstract
This report presents the reproducibility review of the paper “CRDV: Conflict-free Replicated Data Views” [2]. The paper’s artifacts were available and well-documented. The authors provided Docker deployment with all scripts and plotting routines, which made reproducibility straightforward. We successfully reproduced the core claims of the paper and documented any deviations in this report. © 2025 Copyright held by the owner/author(s)

Abstract
This study presents a comprehensive methodology for high-resolution hazard geospatial mapping that combines UAV-GPS surveys and terrestrial LiDAR. This technique uses aerial and ground-based sensing systems to create detailed, three-dimensional terrain models, offering enhanced accuracy and comprehensive coverage. UAVs, combined with high-resolution GPS systems, enable the swift collection of data over large areas, producing detailed orthoimagery and digital surface models. Ground LiDAR systems assess detailed surface features and vertical elements that aerial surveys cannot detect. The combined workflow consists of several stages: (1) equipment preparation and calibration, (2) ground control point deployment, (3) terrestrial LiDAR data acquisition, (4) robust co-registration algorithms for seamless data fusion, (5) data processing and co-registration, (6) quality control and validation and (7) integrated data products. The research validates its findings through field tests across different geological environments, demonstrating that the system achieves centimetre-level precision across various spatial scales. This integrated approach addresses common survey constraints, including vegetation blocking, steepslope shadows and GPS signal loss. Applications of this approach include assessing geohazards, measuring slopes, identifying geostructures, and supporting geological basin analysis. The multisensor framework provides a comprehensive design solution for projects that require broad coverage alongside accurate detail, thereby enhancing the capabilities of geospatial mapping and geovisualisation techniques.

Abstract
Contrastive Analysis (CA) detects anomalies by contrasting patterns unique to a target group (e.g., unhealthy subjects) from those in a background group (e.g., healthy subjects). In the context of brain MRIs, existing CA approaches rely on supervised contrastive learning or variational autoencoders (VAEs) using both healthy and unhealthy data, but such reliance on target samples is challenging in clinical settings. Unsupervised Anomaly Detection (UAD) learns a reference representation of healthy anatomy, eliminating the need for target samples. Deviations from this reference distribution can indicate potential anomalies. In this context, diffusion models have been increasingly adopted in UAD due to their superior performance in image generation compared to VAEs. Nonetheless, precisely reconstructing the anatomy of the brain remains a challenge. In this work, we bridge CA and UAD by reformulating contrastive analysis principles for the unsupervised setting. We propose an unsupervised framework to improve the reconstruction quality by training a self-supervised contrastive encoder on healthy images to extract meaningful anatomical features. These features are used to condition a diffusion model to reconstruct the healthy appearance of a given image, enabling interpretable anomaly localization via pixel-wise comparison. We validate our approach through a proof-of-concept on a facial image dataset and further demonstrate its effectiveness on four brain MRI datasets, outperforming baseline methods in anomaly localization on the NOVA benchmark.

Abstract
This integrative literature review synthesizes insights from multiple disciplines to address the challenges and opportunities in designing digital communication interfaces for persons with Locked-In Syndrome (LIS). The paper highlights the importance of a multidisciplinary approach that includes ethical co-design, visual design principles, and Human-Computer Interaction (HCI). It emphasizes how important it is to have user-friendly, visually appealing, and accessible interfaces to help persons with LIS to communicate more effectively. Important technologies are evaluated for their potential to improve communication, including Augmented and Virtual Reality (AR & VR), Eye Tracking, and Brain-Computer Interfaces (BCI). To guarantee that the emerging technologies are both efficient and considerate of user demands, the review emphasizes the significance of ethical considerations and patient-centered design. This study intends to direct future design-based action research in constructing functional digital communication systems, using head-mounted Extended Reality (XR) technologies, by combining the various research findings from the review.