Model averaging with the hybrid model: An asymptotic study and demonstration

In this paper, we present a new model averaging technique that can be applied in medical research. The dataset is first partitioned by the values of its categorical explanatory variables. Then for each partition, a model average is determined by minimising some form of squared errors, which could be the leave-one-out cross-validation errors. From our asymptotic optimality study and the results of simulations, we demonstrate under several high-level assumptions and modelling conditions that this model averaging procedure may outperform jackknife model averaging, which is a well-established technique. We also present an example where a cross-validation procedure does not work (that is, a zero-valued cross-validation error is obtained) when determining the weights for model averaging.

Elastic anisotropies of deformed upper crustal rocks in the Alps

The crust within collisional orogens is very heterogeneous both in composition and grade of deformation, leading to highly variable physical properties at small scales. This causes difficulties for seismic investigations of tectonic structures at depth since the diverse and partially strong upper crustal anisotropy might overprint the signal of deeper anisotropic structures in the mantle. In this study, we characterize the range of elastic anisotropies of deformed crustal rocks in the Alps. Furthermore, we model average elastic anisotropies of these rocks and their changes with increasing depth due to the closure of microcracks. For that, pre-Alpine upper crustal rocks of the Adula Nappe in the central Alps, which were intensely deformed during the Alpine orogeny, were sampled. The two major rock types found are orthogneisses and paragneisses; however, small lenses of metabasites and marbles also occur. Crystallographic preferred orientations (CPOs) and volume fractions of minerals in the samples were measured using time-of-flight neutron diffraction. Combined with single crystal elastic anisotropies these were used to model seismic properties of the rocks. The sample set shows a wide range of different seismic velocity patterns even within the same lithology, due to the microstructural heterogeneity of the deformed crustal rocks. To approximate an average for these crustal units, we picked common CPO types of rock forming minerals within gneiss samples representing the most common lithology. These data were used to determine an average elastic anisotropy of a typical crustal rock within the Alps. Average mineral volume percentages within the gneiss samples were used for the calculation. In addition, ultrasonic anisotropy measurements of the samples at increasing confining pressures were performed. These measurements as well as the microcrack patterns determined in thin sections were used to model the closure of microcracks in the average sample at increasing depth. Microcracks are closed at approximately 740 MPa yielding average elastic anisotropies of 4 % for the average gneiss. This value is an approximation, which can be used for seismic models at a lithospheric scale. At a crustal or smaller scale, however, local variations in lithology and deformation as displayed by the range of elastic anisotropies within the sample set need to be considered. In addition, larger-scale structural anisotropies such as layering, intrusions and brittle faults have to be included in any crustal-scale seismic model.

The value of Cystatin C in the diagnosis of cardiorenal syndrome type 1

Background The presence of acute kidney injury in the setting of acute heart failure is very common occurrence and was termed cardiorenal syndrome 1 (CRS1). In CRS1 the diagnosis of acute kidney damage is often delayed by creatinine and urine output following KDIGO standards (Kidney Disease Improving Global Outcomes). Cystatin C is one of the earliest markers of worsening renal function. We studied the value of plasma Cystatin C in the diagnosis of cardiorenal syndrome type 1. Purpose This study was aimed: (1) to decribe clinical, subclinical characteristics, prevalence of CRS1; (2) to evaluate the diagnostic efficacy of Cystatin C in diagnosis of CRS1. Materials and method There were 139 patients with acute heart failure or acute decompensated heart failure (ADHF) in the Department of cardiovascular resuscitation and Interventional cardiology at 115 Ho Chi Minh City People's Hospital from September 2018 to June 2019. This is a prospective cohort study. Results There were 48 cases (rate 34.5%) with CRS1, medium age 66.12±15.77, men accounted for 50.4%. The optimal cut-off Cystatin C for diagnosing CRS1 is >1.81 mg/dl, AUC is 0.787 (95% CI 0.71–0.85, p<0.001), sensitivity 75%, specificity 83.52%, positive predictive value 70.6%, negative predictive value 86.4%. Building the optimal regression model by the BMA (Bayesian Model Average) method with only one variable Cystatin C: Odds Ratio = ey, while y = −2.75 + 1.11x Cystatin C. Moreover, a nomogram with variable Cystatin C was designed to predict the likelihood of CRS1 with AUC 0.842. Ultimately, a confusion matrix was constructed to determine model accuracy 81.82%, sensitivity 78.26%, specificity 100%, positive predictive value 100%, negative predictive value 47.37%. Conclusions Cystatin C is quite good value in the diagnosis of CRS1 in patients with acute heart failure or ADHF. A predictive model based on Cystatin C may help early diagnose CRS1 in patients with acute heart failure or ADHF. FUNDunding Acknowledgement Type of funding sources: None.

Simulation for a Mems-Based CTRNN Ultra-Low Power Implementation of Human Activity Recognition

This paper presents an energy-efficient classification framework that performs human activity recognition (HAR). Typically, HAR classification tasks require a computational platform that includes a processor and memory along with sensors and their interfaces, all of which consume significant power. The presented framework employs microelectromechanical systems (MEMS) based Continuous Time Recurrent Neural Network (CTRNN) to perform HAR tasks very efficiently. In a real physical implementation, we show that the MEMS-CTRNN nodes can perform computing while consuming power on a nano-watts scale compared to the micro-watts state-of-the-art hardware. We also confirm that this huge power reduction doesn't come at the expense of reduced performance by evaluating its accuracy to classify the highly cited human activity recognition dataset (HAPT). Our simulation results show that the HAR framework that consists of a training module, and a network of MEMS-based CTRNN nodes, provides HAR classification accuracy for the HAPT that is comparable to traditional CTRNN and other Recurrent Neural Network (RNN) implantations. For example, we show that the MEMS-based CTRNN model average accuracy for the worst-case scenario of not using pre-processing techniques, such as quantization, to classify 5 different activities is 77.94% compared to 78.48% using the traditional CTRNN.

Evaluation of Climate in CMIP6 Models over Two Third Pole Subregions with Contrasting Circulation Systems

Reliable simulations of historical and future climate are critical to assessing ecological and hydrological responses over the Third Pole (TP). In this study, we evaluate the historical and future temperature and precipitation simulations of 18 models from the Coupled Model Intercomparison Project Phase 6 (CMIP6) in southeastern TP (SETP) and the upstream of the Amu Darya and Syr Darya (UAS) regions, two typical TP subregions dominated by the Indian summer monsoon system and westerlies, respectively. Comparison against station observations suggests that CMIP6 models generally capture the intra-annual variability and spatial pattern of historical climate over both subregions. However, the wetting and cold biases observed in CMIP5 still persist in CMIP6; annual temperature is underestimated by most models and annual precipitation is overestimated by all models. Multi-model average cold biases in SETP and UAS are 1.18°C and 0.32°C, respectively, and wet biases in SETP and UAS are 119% and 46%, respectively. We further analyze climate projections under SSP1-2.6, SSP2-4.5, and SSP5-8.5 scenarios. Both SETP and UAS subregions are projected to experience significant warming in 2015–2100, with warming trends 34%–42% and 40%–50% higher than the global trend, respectively. Model projections suggest that the warming trend will slow down under SSP1-2.6 and SSP2-4.5 but further intensify under SSP5-8.5 in 2050–2100. Monsoon-dominated SETP is projected to experience a significant wetting trend stronger than UAS over the entire future period, especially in summer (cf. winter in westerlies-dominated UAS). Concurrently, a significant drying trend in summer is found in UAS during 2050–2100 under SSP5-8.5, suggesting the intensified uneven distributions of seasonal precipitation based on projections.

Enabling Training of Neural Networks on Noisy Hardware

Deep neural networks (DNNs) are typically trained using the conventional stochastic gradient descent (SGD) algorithm. However, SGD performs poorly when applied to train networks on non-ideal analog hardware composed of resistive device arrays with non-symmetric conductance modulation characteristics. Recently we proposed a new algorithm, the Tiki-Taka algorithm, that overcomes this stringent symmetry requirement. Here we build on top of Tiki-Taka and describe a more robust algorithm that further relaxes other stringent hardware requirements. This more robust second version of the Tiki-Taka algorithm (referred to as TTv2) 1. decreases the number of device conductance states requirement from 1000s of states to only 10s of states, 2. increases the noise tolerance to the device conductance modulations by about 100x, and 3. increases the noise tolerance to the matrix-vector multiplication performed by the analog arrays by about 10x. Empirical simulation results show that TTv2 can train various neural networks close to their ideal accuracy even at extremely noisy hardware settings. TTv2 achieves these capabilities by complementing the original Tiki-Taka algorithm with lightweight and low computational complexity digital filtering operations performed outside the analog arrays. Therefore, the implementation cost of TTv2 compared to SGD and Tiki-Taka is minimal, and it maintains the usual power and speed benefits of using analog hardware for training workloads. Here we also show how to extract the neural network from the analog hardware once the training is complete for further model deployment. Similar to Bayesian model averaging, we form analog hardware compatible averages over the neural network weights derived from TTv2 iterates. This model average then can be transferred to another analog or digital hardware with notable improvements in test accuracy, transcending the trained model itself. In short, we describe an end-to-end training and model extraction technique for extremely noisy crossbar-based analog hardware that can be used to accelerate DNN training workloads and match the performance of full-precision SGD.

Analysis and forecasting of crude oil price based on the variable selection-LSTM integrated model

In recent years, the crude oil market has entered a new period of development and the core influence factors of crude oil have also been a change. Thus, we develop a new research framework for core influence factors selection and forecasting. Firstly, this paper assesses and selects core influence factors with the elastic-net regularized generalized linear Model (GLMNET), spike-slab lasso method, and Bayesian model average (BMA). Secondly, the new machine learning method long short-term Memory Network (LSTM) is developed for crude oil price forecasting. Then six different forecasting techniques, random walk (RW), autoregressive integrated moving average models (ARMA), elman neural Networks (ENN), ELM Neural Networks (EL), walvet neural networks (WNN) and generalized regression neural network Models (GRNN) were used to forecast the price. Finally, we compare and analyze the different results with root mean squared error (RMSE), mean absolute percentage error (MAPE), directional symmetry (DS). Our empirical results show that the variable selection-LSTM method outperforms the benchmark methods in both level and directional forecasting accuracy.

Global Rangeland Primary Production and Its Consumption by Livestock in 2000–2010

Livestock grazing occupies ca. 25% of global ice-free land, removing large quantities of carbon (C) from global rangelands (here, including grass- and shrublands). The proportion of total livestock intake that is supplied by grazing (GP) is estimated at >50%, larger than the proportion from crop- and byproduct-derived fodders. Both rangeland productivity and its consumption through grazing are difficult to quantify, as is grazing intensity (GI), the proportion of annual aboveground net primary productivity (ANPP) removed from rangelands by grazing livestock. We develop national or sub-national level estimates of GI and GP for 2000–2010, using remote sensing products, inventory data, and model simulations, and accounting for recent changes in livestock intake, fodder losses and waste, and national cropland use intensities. Over the 11 study years, multi-model average global rangeland ANPP varied between the values of 13.0 Pg C in 2002 and 13.96 Pg C in 2000. The global requirement for grazing intake increased monotonically by 18%, from 1.54 in 2000 to 1.82 Pg C in 2010. Although total global rangeland ANPP is roughly an order of magnitude larger than grazing demand, much of this total ANPP is unavailable for grazing, and national or sub-national deficits between intake requirements and available rangeland ANPP occurred in each year, totaling 36.6 Tg C (2.4% of total grazing intake requirement) in 2000, and an unprecedented 77.8 Tg C (4.3% of global grazing intake requirement) in 2010. After accounting for these deficits, global average GI ranged from 10.7% in 2000 to 12.6% in 2009 and 2010. The annually increasing grazing deficits suggest that rangelands are under significant pressure to accommodate rising grazing demand. Greater focus on observing, understanding, and managing the role of rangelands in feeding livestock, providing ecosystem services, and as part of the global C cycle, is warranted.

Analysis of Cargo Airline Services Quality in Cargo Delivery In Nigeria

This study analysed cargo airline services and cargo delivery in Nigeria. The focus was on identifying the disparity between expected and perceived airline service quality on cargo delivery. This study was carried out in Lagos State, Nigeria. Random sampling technique was used to select 239 freight forwarders agents that operate in the Cargo Terminal of Muritatal Muhammed International Airport, Lagos State. SERVQUAL model and t-test were used to analyse the disparity between expectations and perceptions of airline service quality. SERVQUAL model average gap score is -0.126 which implies that customers are not satisfied with cargo airline service quality. Moreover, the findings revealed that there is a negative disparity between customer expectations and perceptions of airline service quality on air cargo delivery with Reliability having the lowest negative t-value of (-34.409), followed by Tangibles with a t-value of (-23.300), Responsiveness has a t-value of (-12.910), Assurance followed with t-value of (-13.572) while Empathy has a t-value of (-10.165). Conclusion is drawn that cargo airline customers' expectations about their service quality are more than they really perceived. Recommendation was made that cargo airline should make an assessment of service quality dimensions (Tangibles, Reliability, Responsiveness, Assurance and Empathy) and provide a means to improve upon them.