Vision Transformers (ViTs) have revolutionized the field of computer vision, yet their deployments on resource-constrained devices remain challenging due to high computational demands. To expedite ...pre-trained ViTs, token pruning and token merging approaches have been developed, which aim at reducing the number of tokens involved in the computation. However, these methods still have some limitations, such as image information loss from pruned tokens and inefficiency in the token-matching process. In this paper, we introduce a novel Graph-based Token Propagation (GTP) method to resolve the challenge of balancing model efficiency and information preservation for efficient ViTs. Inspired by graph summarization algorithms, GTP meticulously propagates less significant tokens' information to spatially and semantically connected tokens that are of greater importance. Consequently, the remaining few tokens serve as a summarization of the entire token graph, allowing the method to reduce computational complexity while preserving essential information of eliminated tokens. Combined with an innovative token selection strategy, GTP can efficiently identify image tokens to be propagated. Extensive experiments have validated GTP's effectiveness, demonstrating both efficiency and performance improvements. Specifically, GTP decreases the computational complexity of both DeiT-S and DeiT-B by up to 26% with only a minimal 0.3% accuracy drop on ImageNet-1K without finetuning, and remarkably surpasses the state-of-the-art token merging method on various backbones at an even faster inference speed. The source code is available at https://github.com/Ackesnal/GTP-ViT.
The Putonghua Proficiency Test (Putonghua Shuiping Ceshi, PSC) is a speaking test in China. The propositional speaking section in PSC focuses on the ability of speakers to express ideas fluently and ...accurately without textual reference. However, unlike other sections of the PSC, propositional speaking is still scored manually, which can result in inefficiency, high costs, and subjectivity. To address these issues, an automatic speech fluency evaluation method based on multimodality is proposed. First, different neural networks are used to extract unimodal features. Then, cross-modal attention is applied to achieve multimodal fusion. Finally, fluency evaluation results are obtained using self-attention to reinforce high-contributing information. The proposed method achieves 81.67% accuracy on a self-built dataset, demonstrating that combining textual and acoustic features provides complementary information to improve automatic speech fluency evaluation accuracy.
Identifying abnormalities from neuroimaging of brain matters has been a crucial way of diagnosis of two closely associated diseases, namely Alzheimer׳s Disease (AD) and Mild Cognitive Impairment ...(MCI). Different types of neuroimaging have been developed to help such diagnosis, and significant research efforts are put into the automation and quantification of such diagnosis by computer algorithms over the past decades. In this paper we propose an ensemble learning framework to create effective models for AD/MCI related classification tasks from multiple modalities of neuroimaging and multiple baseline estimators. The framework is based on artificial neural networks and it resembles a composite model that solves the feature fusion learning problem as well as the prediction problem simultaneously, which targets at exploiting the prediction power of both fusing multiple data modalities and leveraging multiple mutually complementary classification models. We conduct extensive experiments on the well-known ADNI dataset and find that the proposed model works demonstrate advantages for both of the classification tasks studied.
Full text
Available for:
GEOZS, IJS, IMTLJ, KILJ, KISLJ, NUK, OILJ, PNG, SAZU, SBCE, SBJE, UL, UM, UPCLJ, UPUK, ZRSKP
This case study draws a specific link to the practice of action learning (AL) in China. We organized ourselves into an AL set and used Revans' AL, as interpreted by Marquardt (
2004
), to create a ...post-teaching dialog to examine the experience gained from delivering an interdisciplinary course online-a novice situation-in a Chinese transnational university. AL's questioning insight occurred after conducting an evidence-based evaluation of online teaching in an interdisciplinary higher education course that used Debattista's (
2018
) online teaching effectiveness rubric. The rubric offered rich 'programed knowledge' that triggered our question-based inquiry. We conclude that our AL approach is valuable for teacher professional development and offers our rationale for why this particular AL practice would be suitable for the Confucian culture, teaching of interdisciplinary courses, and in novice situations. By conducting AL, we identified a list of key findings, such as proactive communication among teachers in an interdisciplinary course, was vital when teaching to a large group of students. We offer recommendations to improve interdisciplinary online course design and delivery in the future based on reflections from the AL. Implications, limitations, and suggestions for future research are presented at the end of this paper.
Full text
Available for:
BFBNIB, NUK, PILJ, SAZU, UL, UM, UPUK
Experiments on the Sb-S-Se system were conducted at 300°C, and a continuous stibnite-antimonselite binary solid solution was established. By substituting S for Se, the compositions of S-rich and ...Se-rich endmembers were confirmed as Sb
2
S
3
and Sb
2
Se
3
, respectively. Based on Se/(S+Se) ratios of microprobe analyses, binary stibnite-antimonselite solid solutions are defined as stibnite, selenium stibnite, sulfur antimonselite, and antimonselite. Microhardness of the stibnite subseries (Sb = 60.11-72.58, S = 13.20-27.63, and Se = 0.00-27.23 wt%) and the antimonselite subseries (Sb = 49.29-59.25, Se = 28.89-51.94, S = 0.00-12.10 wt%) varies from 112.95 to 127.72 kg/mm
2
. The variation of Se concentration is continuous throughout the series, confirming a random substitution of Se for S. Crystallographic parameters obtained from the series vary as follows: a = 1.123375-1.163890 nm, b = 1.132502-1.179553 nm, c = 0.383914-0.398071 nm, D = 4.593-5.896 g.cm
-3
, and V = 0.488425-0.546500 nm
3
. As evident from the above data, the higher the Se concentration, the larger the crystallographic parameters. The Sb-S-Se binary solid solutions obey Vegard's law.
Full text
Available for:
BFBNIB, GIS, IJS, KISLJ, NUK, PNG, UL, UM, UPUK
Vancomycin usage is often unavoidable in pregnant patients; however, literature suggests vancomycin can cross the placental barrier and reach the fetus. Understanding the mass transit of vancomycin ...to the fetus is important in pregnancy. We aimed to (i) identify a relevant population pharmacokinetic (PK) model for vancomycin in pregnancy and (ii) estimate PK parameters and describe the mass transit of vancomycin from mother to pup kidneys. Pregnant Sprague-Dawley rats (i.e., trimester 1 and trimester 3) received 250 mg/kg vancomycin once daily for three days through intravenous injection via an internal jugular vein catheter. Vancomycin concentrations in maternal plasma and pup kidneys were quantified via liquid chromatography-tandem mass spectrometry (LC-MS/MS). Multiple compartment models were fitted and assessed using a nonparametric approach with Pmetrics. A total of 10 vancomycin-treated rats and 48 pups contributed PK data. A 3-compartment model adjusted for trimester fit the data well (maternal plasma Bayesian, observed versus predicted R
= 0.978; pup kidney Bayesian, observed versus predicted R
= 0.999). The mean rate constant for vancomycin mass transit to the pup kidney was 0.72 h
for trimester 1 dams and 0.75 h
for trimester 3 dams. Median vancomycin concentrations in pup kidneys from trimester 3 were significantly higher than those in trimester 1 (8.62 versus 0.36 μg/mL,
< 0.001). Vancomycin transited to the fetus from the mother and was; kidney accumulation differed by trimester. This model may be useful for a translational understanding of vancomycin distribution in pregnancy to ensure efficacious and safe doses to both mother and fetus.
Qinling orogenic belt hosts many porphyry Mo deposits, which were formed by ore-forming fluids exsolved from oxidized magmas. Detailed processes of ore fluids evolution and metal deposition ...mechanisms of porphyry Mo deposits remain unclear. The quartz solubility model and cathodoluminescence (CL) images combined with fluid inclusions (FIs) and isotope analyses at the Nantai Mo deposit provide new insights into processes and physicochemical conditions for Mo mineralization. Based on field and microscopy of ore and alteration assemblages together with CL images, four types of hydrothermal veins (i.e., A, B, C, and D veins) have been identified. The quartz in A veins intergrown with K-feldspar precipitated from high-temperature and low- to intermediate-salinity fluids under near-lithostatically conditions. B vein comprises quartz and K-feldspar and forms at 336–466 °C, pH from 6.30 to 11.02, and oxygen fugacity (fO2) of −30.81 to −24.26. Hydrothermal fluids during this stage underwent phase separation under ductile to brittle transition. C veins include muscovite and most molybdenite without quartz, indicating that they formed in the retrograde quartz solubility field at 353 to 398 °C. Thermodynamic calculation suggests molybdenite mainly precipitated under fO2 from −32.37 to −29.05 and pH from 2.68 to 6.54. Quartz in D veins formed from low-temperature and low-salinity fluids. Sulfur isotopic analyses suggest that the ore components are derived from the old crust. The main factors controlling Mo mineralization at Nantai are the decreases in fO2 and pH values of hydrothermal fluids.
Display omitted
•The ore-forming components were sourced from magma that derived from an old crust.•The mark of ore-forming fluids is high-temperature, low- to intermediate-salinity, and intermediate-density.•The decrease in fO2 and pH of the hydrothermal fluids lead to molybdenite formation.
Full text
Available for:
GEOZS, IJS, IMTLJ, KILJ, KISLJ, NLZOH, NUK, OILJ, PNG, SAZU, SBCE, SBJE, UILJ, UL, UM, UPCLJ, UPUK, ZAGLJ, ZRSKP
The ore-forming material sources of the Baiyangping copper-cobalt-silver polymetallic deposit have been studied in view of the S, Pb, C, O and H isotopic characteristics and the ratio of Co/Ni of ...cobaltite. The results showed that sulfur in metallic sulfides may have come from a mixed sulfur-source consisting of the sulfur-source from metamorphic rocks in the basin basement with basic volcanic rocks and the sulfur-source from basin sulfates; lead in the ores was provided by the sedimentary rocks and basement rocks; CO2 in ore-forming fluids was derived from thermolysis of altered and normal marine facies carbonates and decarboxylation of sedimentary organic matter respectively; the ore-forming fluids belong to the SO4-Cl-Na-Ca-type basin thermal brines derived from paleo-meteoric waters; cobalt in the deposit may also be derived from the metamorphic rocks in the basin basement with basic volcanic rocks.
Full text
Available for:
DOBA, IZUM, KILJ, NUK, PILJ, PNG, SAZU, SIK, UILJ, UKNU, UL, UM, UPUK
Evidence for open-system magmatic processes related to wallrock assimilation accompanied by fractional crystallization (AFC) is present in the Guangshigou biotite pegmatites, North Qinling Orogen. ...The biotite pegmatite-gneiss contacts generally coincide with the greatest enrichment of U and Th. Zircon UPb dating constrains the crystallization ages of the biotite pegmatite (rim zone-415±2.6Ma; internal zone-413.5±2.5Ma), in line with a pyrite PbPb isochron age (413±22Ma). Metamict areas in zircon show generally elevated concentrations of trace elements and expulsion of radioactive Pb. Internal zone samples, representing uncontaminated magma, have negative to positive zircon (~413Ma) εHf(t) (−1.53−+3.24), low εNd(t) values (−2.4), and old Hf and Nd model ages (tDM2=1.5–1.19Ga, T2DM=1.35Ga, respectively), indicating a dominantly recycled Mesoproterozoic lower crustal material with involvement of some juvenile materials in the source region. The magmatic oxygen fugacity (fO2) and crystallization temperatures ranges from −24.81 to −13.34 of log fO2 and 570°C to 793°C, respectively. Compared to the internal zone, pegmatite rim samples display a variable and lower εNd(t) values (−3.9 to −2.8) and T2DM (1.47–1.37Ga), but similar Hf isotopic compositions, favouring a three-component isotopic mixing model (recycled Mesoproterozoic lower crust materials, juvenile materials, and host gneiss). Pronounced variations of Ti, Y, U, Th, Hf, and REE concentrations in zircon from grain to grain in individual samples and from area to area within individual grains suggest a fluctuating crystallization environment in hybridized magma from which the rim-hosted zircons crystallized. Variable and high radiogenic Pb ratios of pyrites forming in the hybridized magma were inherited from the matrix. Zircons from both zones exhibit similar Hf isotope patterns, indicating the rim-hosted zircons crystallized during the early stage of hybridization of magma. However, the heterogeneous Nd isotopic composition in individual pegmatites imply an incorporation of wallrock-derived melt with less radiogenic Nd.
•Heterogeneous chemical compositions of individual biotite pegmatites arise from assimilation-fractional crystallization.•Elevated trace element concentrations and expulsion of radioactive Pb occur in zircon metamict areas.•Effect of contamination of pegmatite magma by wallrock on zircon LuHf system of the magma was insignificant.•Xenolithic melt with less radiogenic Nd is mixed into the host pegmatite magma.
Full text
Available for:
GEOZS, IJS, IMTLJ, KILJ, KISLJ, NLZOH, NUK, OILJ, PNG, SAZU, SBCE, SBJE, UILJ, UL, UM, UPCLJ, UPUK, ZAGLJ, ZRSKP
Increasing production personalization demand requires manufacturing enterprises to gain higher flexibility and faster market response. Enterprises have to share their manufacturing resources and ...cooperate with each other to win out in the fierce market competition. Under the circumstances, Social Manufacturing (SocialM) mode is proposed. In this mode, Social Manufacturing Network (SMN) integrates distributed Socialized Manufacturing Resources (SMRs) to provide more precise and professional service for customers. As a decentralized network, SMN cannot ensure the cross-enterprise collaborations because there is no trusted third party as supervisor. In this paper, a blockchain-based Production Credit Mechanism (PCM) for manufacturing services is put forward to regulate the cross-enterprise collaborations among SMRs. The frame and concept of PCM are firstly given, followed by four key enabling technologies supporting the implementation of the mechanism. It is expected that the PCM proposed in this paper will provide a possible way to normalize and regulate cross-enterprise collaborations under social manufacturing mode.