No abstract

No abstract

No abstract

No abstract

Soil, water, and air NO pollution is a major environmental problem worldwide. Stable isotope analysis can assess the origin of NO because different NO sources carry different isotope signatures. Hence, using appropriate chemical methods to determine the δN-NO values in different samples is important to improve our understanding of the N-NO pollution and take possible strategies to manage it. Two modified chemical methods, the cadmium-sodium azide method and the VCl-sodium azide method, were used to establish a comprehensive inventory of δN-NO values associated with major NO fluxes by the conversion of NO into NO. Precision and limit of detection values demonstrated the robustness of these quantitative techniques for measuring δN-NO. The standard deviations of the δN-NO values were 0.35 and 0.34‰ for the cadmium-sodium azide and VCl-sodium azide methods. The mean δN-NO values of river water, soil extracts, and summer rain were 8.9 ± 3.3, 3.5 ± 3.5, and 3.3 ± 2.1‰, respectively. The δN-NO values of low concentration samples collected from coal-fired power plants, motor vehicles, and gaseous HNO was 20.3 ± 4.3, 5.6 ± 2.78, and 5.7 ± 3.6‰, respectively. There was a good correlation between the δN-NO compositions of standards and samples, which demonstrates that these chemical reactions can be used successfully to assess δN-NO values in the environment.

Mafic dyke swarms are excellent time markers and paleostress indicators. Numerous late Paleoproterozoic mafic dykes are exposed throughout the Trans-North China Orogen (TNCO). Most of these dykes trend NW-SE or NNW-SSE, nearly parallel to the orogen, while a series of E-W trending mafic dykes are restricted in the L€ uliang and southern Taihang areas in the central segment of the TNCO. These dykes were mostly considered to be linked with breakup of the supercontinent Columbia previously. In this study, 16 mafic dykes were investigated in the L€ uliang Complex. Zircon LA-ICP-MS dating of four samples yields magmatic crystallization ages of 1.78-1.79 Ga. These dykes belong to the tholeiite series and consist of basalt, basaltic andesite, and andesite. They are enriched in LREE and LILE and depleted in HFSE, and have negative zircon eHf (t) values of 21.7 to 212.2. The E-W trending mafic dykes show similar geochemical and isotopic features compared to the NW-SE trending dykes in other complexes. They were most likely originated from a lithospheric mantle metasomatized by subduction-related fluids and later emplaced along extensional fractures in a postcollisional setting. NW-SE trending fractures were formed due to gravitational collapse and thinning of the lithosphere. E-W trending fractures in the central segment of the orogen constitute a transverse accommodation belt to equilibrate the different amounts of extension between the northern and southern TNCO. The impact of the postorogenic extension might have continued to approximately 1680 Ma as evidenced by the presence of abundant approximately 1750-1680 Ma anorthositegabbro-mangerite-rapakivi granite suites (AMCG-like) occurring in the northern NCC.

Although more layers and more parameters generally improve the accuracy of the models, such big models generally have high computational complexity and require big memory, which exceed the capacity of small devices for inference and incurs long training time. In addition, it is difficult to afford long training time and inference time of big models even in high performance servers, as well. As an efficient approach to compress a large deep model (a teacher model) to a compact model (a student model), knowledge distillation emerges as a promising approach to deal with the big models. Existing knowledge distillation methods cannot exploit the elastic available computing resources and correspond to low efficiency. In this paper, we propose an Elastic Deep Learning framework for knowledge Distillation, that is, EDL‐Dist. The advantages of EDL‐Dist are threefold. First, the inference and the training process is separated. Second, elastic available computing resources can be utilized to improve the efficiency. Third, fault‐tolerance of the training and inference processes is supported. We take extensive experimentation to show that the throughput of EDL‐Dist is up to 3.125 times faster than the baseline method (online knowledge distillation) while the accuracy is similar or higher.

No abstract