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 over NVIDIA V100 and A100 GPUs show that MinatoLoader accelerates training by up to 7.5× (3.6× on average) over PyTorch DataLoader and Pecan, and up to 3× (2.2× on average) over DALI. It also increases average GPU utilization from 46% with PyTorch to 90%, while preserving model accuracy and enabling faster convergence. © 2026 Copyright held by the owner/author(s)

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.

Abstract
Introduction Immersive virtual reality (iVR) offers a multisensory environment for education, yet the integration of olfaction remains underexplored. This study examined whether incorporating ambient olfactory stimuli into an iVR environment enhances foreign language vocabulary retention and the user's sense of presence.Methods A between-subjects experiment was conducted with 59 participants who learned German vocabulary in a virtual airport scenario. Participants were assigned to one of five ambient olfactory conditions systematically selected to represent distinct quadrants of the circumplex model of affect: no scent (control), spearmint (pleasant-arousing), lavender (pleasant-calming), burning wood (unpleasant-arousing), or sewage (unpleasant-calming). Vocabulary retention was measured using matching pre- and post-tests, while subjective presence was assessed using the standardised Igroup Presence Questionnaire (IPQp).Results The results indicated that ambient olfactory stimulation, regardless of affective valence or arousal level, did not significantly improve immediate vocabulary retention compared to the control condition. However, scent did impact the subjective experience of presence; notably, an unpleasant, high-arousal scent (burning wood) served as a distraction, significantly reducing perceived spatial presence.Discussion These findings establish an important boundary condition for multisensory educational VR. They demonstrate that the simple addition of ambient, affective scents as a background stimulus is insufficient to drive immediate cognitive learning gains, and may even detract from immersion if unpleasant. Multisensory iVR design must be guided by pedagogical priorities rather than novelty alone, suggesting that relying solely on ambient emotional modulation via olfaction is not a viable strategy for complex cognitive tasks.

Abstract
We consider the problem of selecting bases for firefighting activities (e.g., vigilance, water refill, initial attack) and links between them in the context of wildfire promptness. Bases can be facilities, such as watchtowers and water tanks, or positions from where an initial attack is conducted. It is assumed that it is advantageous to connect bases in such a way that resources (e.g. ground crews) can quickly move between them. The general problem is modelled in a general way as integration of a set covering problem (for selecting the location of the bases) and a travelling salesman problem where the cities are the selected locations and the arcs the links that connect them. We propose a mixed integer programming model where objectives are addressed by lexicographic optimization. The first objective is related to cover potential ignition points with a high estimate of their initial spread rate of the fire at the detection time. Computational experiments are discussed for a scenario, of an actual landscape, with parameters estimated from a fire behaviour model that takes into account slope, fuels, and wind. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2026.

Abstract
Shipping container unloading is a physically demanding task often carried out under challenging conditions, which motivates the use of automation. However, automating this process is complex due to the unpredictable sizes and quantities of each shipment. Existing solutions tend to be task-specific, rely on closed software stacks, and offer limited information on performance in non-controlled environments, which restricts their adaptability. We present CARGO, a modular pipeline that enables a mobile manipulator equipped with regular sensors and actuators to unload containers autonomously. The pipeline employs a predefined, layered workflow composed of reconfigurable modules that can be adapted to various robots, ensuring that all boxes in a stack are systematically handled. In simulation, the pipeline successfully unloaded a full container without collisions, thereby validating the complete workflow. Laboratory tests further confirmed these results, with the mobile manipulator successfully unloading boxes across multiple trials, with a success rate of 97%. These results demonstrate that a versatile mobile manipulator can handle mixed box sizes and chaotic layouts using a generic, modular pipeline, highlighting a promising direction for flexible container-unloading automation. © 2026 IEEE.