Search

Distributional shift, or the mismatch between training and deployment data, is a significant obstacle to the usage of machine learning in high-stakes industrial applications, such as autonomous driving and medicine. This creates a need to be able to assess how robustly ML models generalize as well as the quality of their uncertainty estimates. Standard ML baseline datasets do not allow these properties to be assessed, as the training, validation and test data are often identically distributed. Recently, a range of dedicated benchmarks have appeared, featuring both distributionally matched and shifted data. Among these benchmarks, the Shifts dataset stands out in terms of the diversity of tasks as well as the data modalities it features. While most of the benchmarks are heavily dominated by 2D image classification tasks, Shifts contains tabular weather forecasting, machine translation, and vehicle motion prediction tasks. This enables the robustness properties of models to be assessed on a diverse set of industrial-scale tasks and either universal or directly applicable task-specific conclusions to be reached. In this paper, we extend the Shifts Dataset with two datasets sourced from industrial, high-risk applications of high societal importance. Specifically, we consider the tasks of segmentation of white matter Multiple Sclerosis lesions in 3D magnetic resonance brain images and the estimation of power consumption in marine cargo vessels. Both tasks feature ubiquitous distributional shifts and a strict safety requirement due to the high cost of errors. These new datasets will allow researchers to further explore robust generalization and uncertainty estimation in new situations. In this work, we provide a description of the dataset and baseline results for both tasks.

The problem of missing data, usually absent incurated and competition-standard datasets, is an unfortunate reality for most machine learning models used in industry applications. Recent work has focused on understanding the nature and the negative effects of such phenomena, while devising solutions for optimal imputation of the missing data, using both discriminative and generative approaches. We propose a novel mechanism based on multi-head attention which can be applied effortlessly in any model and achieves better downstream performance without the introduction of the full dataset in any part of the modeling pipeline. Our method inductively models patterns of missingness in the input data in order to increase the performance of the downstream task. Finally, after evaluating our method against baselines for a number of datasets, we found performance gains that tend to be larger in scenarios of high missingness., Comment: 8 pages, 7 figures

Highlights • In water-rich silicic systems effusions can be generated by magmas with higher water contents than explosions. • The paper presents the comparative example of effusive and explosive deposits erupted at Nisyros-Yali. • Explosive eruptions were generated by water-rich magmas that were water-undersaturated during pre-eruptive storage. • Effusive eruptions were generated by water-richer magmas that were water-supersaturated during pre-eruptive storage. • The role of pre-eruptive exsolved volatiles in controlling eruptive styles is explored. Arc volcanoes generally emit water-rich, high-viscosity silicic magmas, which are prone to erupt explosively. However, effusive behavior is a common occurrence despite the high-H2O, high viscosity conditions. The contrasting shift from effusive to explosive behavior (and vice-versa) at any individual volcano raises the question on what controls eruptive style. Permeability development in conduits allows magma to outgas and is clearly a key factor. However, an important question is whether magma reservoir processes can also have an influence on eruptive styles. The answer could have direct impact on predicting eruptive behavior. Hence, we explore this potential connection by analyzing nine alternating effusive and explosive silicic deposits that were emplaced during distinct eruptions at the active Nisyros-Yali volcanic center. The lavas and pyroclastic deposits are compositionally similar. This indicates a negligible influence of the bulk rock composition on different eruptive styles. The crystal contents vary between units, without any clear correlation with eruptive style (from nearly aphyric to ~45 vol% crystals). Mineral textures and chemistry do show variations between effusive and explosive eruptions, with a larger proportion of resorbed plagioclase and, in most cases, more evolved amphiboles present in the lava flows. Mineral thermo-barometry and hygrometry show that the storage zones of magmas generating effusive

The fumarolic mineralogy of the Icelandic active volcanoes, the Tyrrhenian volcanic belt (Italy) and the Aegean active arc (Greece) is investigated, and literature data surveyed in order to define the characteristics of the European fumarolic systems. They show broad diversity of mineral associations, with Vesuvius and Vulcano being also among the world localities richest in mineral species. Volcanic systems, which show recession over a longer period, show fumarolic development from the hightemperature alkaline halide/sulphate, calcic sulphate or sulphidic parageneses, synchronous with or immediately following the eruptions, through mediumtemperature ammonium minerals, metal chlorides, or fluoride associations to the late low-temperature paragenesis dominated by sulphur, gypsum, alunogen, and other hydrous sulphates. The situation can be different in the systems that are not recessing but show fluctuations in activity, illustrated by the example of Vulcano where the high-temperature association appears intermittently. A full survey of the mineral groups and species is given in respect to their importance and appearance in fumarolic associations.

There exist several scheduling schemes for parallelizing loops without dependences for shared and distributed memory systems. However, efficiently parallelizing loops with dependences is a more complicated task. This becomes even more difficult when the loops are executed on a distributed memory cluster where communication and synchronization can be a bottleneck. The problem lies in the processor idle time which occurs during the beginning and final stages of the execution. In this paper we propose a new scheduling scheme that minimizes the processor idle time and thus it enhances load balancing and performance. The new scheme is applied to two-dimensional iteration spaces with dependences. The proposed scheduling scheme follows a tiled wavefront pattern in which the tile size gradually decreases in all dimensions. We have tested the proposed scheme on a dedicated and homogeneous cluster of workstations and we verified that it significantly improves execution times over scheduling using traditional tiling. © 2007 IEEE.

In recent content delivery mechanisms, popular contents tend to be placed closer to the users for better delivery performance and lower network resource occupation. Caching mechanisms in Content Delivery Networks (CDN), Mobile Edge Clouds (MECs) and fog computing have implemented edge caching paradigm for different application scenarios. However, state-of-the-art caching mechanisms in literature are mostly bounded by application scenarios. With the rapid development of heterogeneous networks, the lack of uniform caching management has become an issue. Therefore, a novel caching mechanism, Semi-Edge caching (SE), is proposed in this paper. SE caching mechanism is based on in-network caching technique and it could be generically applied into various types of network fog. Furthermore, two content allocation strategies, SE-U (unicast) and SE-B (broadcast), are proposed within SE mechanism. The performance of SE-U and SE-B are evaluated in three typical topologies with various scenario contexts. Compared to edge caching, SE can reduce latency by 7% and increase cache hit ratio by 45%.

The rapid growth in computation and processing power of end-user devices has transitioned network functionalities from the core network to the Fog, thus, reducing response times and ultimately improving user experience. The Information-Centric Networking paradigm aims to transform conventional content caching and delivery approaches by enabling Fog nodes to participate in both forwarding and caching. In this work, a Fog caching design scheme is presented for applications such as Internet-of-Things, which integrates three novel design attributes. Firstly, a Fog cluster-based scheme is proposed that utilises both in-network and end-user devices to cache content closer to the edge network, according to the increasing popularity of the content. Secondly, this work proposes a near-path approach that leverages caching nodes near the content delivery path. Finally, by craftily integrating reactive and proactive caching, congestion during network peak-time is averted. Simulations are conducted in the Icarus environment and evaluated against eight popular benchmark techniques. The results indicate significant improvement in internal link load and path stretch metrics. Finally, the cache hit ratio metric is consistently better than all other benchmarks, while the latency performance of the proposed scheme is competitive when the content distribution is more concentrated.