Distributed Computing and Inference for Big Data

In this study, we examined whether age can be predicted on the basis of different anatomical features obtained from a large sample of healthy subjects (n = 3,144). From this sample we obtained different anatomical feature sets: (1) 11 larger brain regions (including cortical volume, thickness, area, subcortical volume, cerebellar volume, etc.), (2) 148 cortical compartmental thickness measures, (3) 148 cortical compartmental area measures, (4) 148 cortical compartmental volume measures, and (5) a combination of the above-mentioned measures. With these anatomical feature sets, we predicted age using 6 statistical techniques (multiple linear regression, ridge regression, neural network, k-nearest neighbourhood, support vector machine, and random forest). We obtained very good age prediction accuracies, with the highest accuracy being R2 = 0.84 (prediction on the basis of a neural network and support vector machine approaches for the entire data set) and the lowest being R2 = 0.40 (prediction on the basis of a k-nearest neighborhood for cortical surface measures). Interestingly, the easy-to-calculate multiple linear regression approach with the 11 large brain compartments resulted in a very good prediction accuracy (R2 = 0.73), whereas the application of the neural network approach for this data set revealed very good age prediction accuracy (R2 = 0.83). Taken together, these results demonstrate that age can be predicted well on the basis of anatomical measures. The neural network approach turned out to be the approach with the best results. In addition, it was evident that good prediction accuracies can be achieved using a small but nevertheless age-representative dataset of brain features. Hum Brain Mapp 38:997-1008, 2017. © 2016 Wiley Periodicals, Inc.

The multivariate probit is popular for modeling correlated binary data, with an attractive balance of flexibility and simplicity. However, considerable challenges remain in computation and in devising a clear statistical framework. Interest in the multivariate probit has increased in recent years. Current applications include genomics and precision medicine, where simultaneous modeling of multiple traits may be of interest, and computational efficiency is an important consideration. We propose a fast method for multivariate probit estimation via a two-stage composite likelihood. We explore computational and statistical efficiency, and note that the approach sets the stage for extensions beyond the purely binary setting.

We present a novel statistical approach to incorporate uncertainty awareness in model-free distributional deep reinforcement learning for mission and safety-critical robotics. Deep learning predictions are influenced by uncertainties in the data, termed as aleatoric uncertainties, as well as uncertainties in the learning process and model structure, known as epistemic uncertainties. The proposed algorithm, called as Calibrated Evidential Quantile Regression in Deep-Q Networks (CEQR-DQN), addresses key challenges associated with separately estimating aleatoric and epistemic uncertainty in stochastic robotic environments. It combines deep evidential learning with quantile calibration based on the principles of conformal inference to provide explicit, sample-free computations of global uncertainty as opposed to local estimates based on simple variance. Thereby, the proposed approach overcomes limitations of traditional methods in computational and statistical efficiency and handling of out-of-distribution (OOD) observations. Tested on a suite of representative miniaturized Atari games (i.e., MinAtar), CEQR-DQN is shown to surpass similar existing frameworks in scores and learning speed. Its ability to rigorously evaluate uncertainties improves exploration strategies and can serve as a blueprint for other uncertainty-aware robotic algorithms.

Hantush’s model is widely used for predicting rise in water table in response to groundwater recharge. Several approximate methods of the Hantush mound function, S(α, β) have been developed to overcome the limitations of Hantush’s tabulated values of the S(α, β) function. These approximate methods have their own advantages and disadvantages, and it is difficult to identify the most accurate and computationally efficient S(α, β) estimation method. In this study, performance of four different algebraic approximate S(α, β) estimation methods are compared with the Hantush method using the published data. The four different methods considered are Swamee and Ohja (1997) (SO), Singh (2012) (SI), Vatankhah (2013) (VA) and Gauss-Legendre quadrature (GL) method with various Gaussian points (GP). Seven statistical accuracy and computation efficiency indicators are used to assess the performance of different S(α, β) estimation methods. The GL method with 100 to16 GPs is found to be the most accurate S(α, β) estimation method. This is followed by the SO, GL with 14 to 12 GPs, VA, GL with 10 GP, SI, and GL with 9 to 3 GPs. A good trade off between accuracy and efficiency is found with the SO, VA, and GL method with 14, 12 and 10 GPs. Comprehensive analysis of different S(α, β) estimation methods, and their ranking based on overall performance index will be helpful in modelling water table rise due to groundwater recharge, optimum design of recharge basin, and evaluation of effectiveness of recharge basins in groundwater recharging.

Distributed learning for Random Vector Functional-Link networks

Deep residual networks (ResNets) have shown state-of-the-art performance in various real-world applications. Recently, the ResNets model was reparameterized and interpreted as solutions to a continuous ordinary differential equation or Neural-ODE model. In this study, we propose a neural generalized ordinary differential equation (Neural-GODE) model with layer-varying parameters to further extend the Neural-ODE to approximate the discrete ResNets. Specifically, we use nonparametric B-spline functions to parameterize the Neural-GODE so that the trade-off between the model complexity and computational efficiency can be easily balanced. It is demonstrated that ResNets and Neural-ODE models are special cases of the proposed Neural-GODE model. Based on two benchmark datasets, MNIST and CIFAR-10, we show that the layer-varying Neural-GODE is more flexible and general than the standard Neural-ODE. Furthermore, the Neural-GODE enjoys the computational and memory benefits while performing comparably to ResNets in prediction accuracy.

In this paper, a distributed learning leader-follower consensus protocol based on Gaussian process regression for a class of nonlinear multi-agent systems with unknown dynamics is designed. We propose a distributed learning approach to predict the residual dynamics for each agent. The stability of the consensus protocol using the data-driven model of the dynamics is shown via Lyapunov analysis. The followers ultimately synchronize to the leader with guaranteed error bounds by applying the proposed control law with a high probability. The effectiveness and the applicability of the developed protocol are demonstrated by simulation examples.

Ship dynamic models serve as the foundation for designing ship controllers, trajectory planning, and obstacle avoidance. Support vector regression (SVR) is a commonly used nonparametric modelling method for ship dynamics. Achieving high accuracy SVR models requires a substantial amount of training samples. Additionally, as the number of training samples increases, the computational efficiency for solving the quadratic programming problem (QPP) of SVR decreases. Ship controllers demand dynamic models with both high accuracy and computational efficiency. Therefore, to enhance the prediction accuracy and computational efficiency of SVR, this paper proposes a nonparametric modelling method based on twin SVR (TSVR). TSVR replaces a large QPP with a set of smaller QPPs, significantly enhancing generalizability and computational efficiency. To further improve the predictive accuracy of TSVR, the puma optimizer algorithm is employed to determine the optimal hyperparameters. The performance of the proposed method is validated using a Mariner class vessel. Gaussian white noise is introduced into the modelling data to simulate measurement error. The TSVR model accurately predicts various zigzag and turning circle manoeuvring motions under disturbance conditions, demonstrating its robustness and generalizability. Compared to the SVR model, the TSVR model achieves lower root mean square error and computational time, confirming its superior predictive accuracy and computational efficiency.

A comparison of alternative methods for the estimation of dynamic factor demand models under non-static expectations

In this paper, we introduce Traversal Learning (TL), a novel approach designed to address the problem of decreased quality encountered in popular distributed learning (DL) paradigms such as Federated Learning (FL), Split Learning (SL) and SplitFed Learning (SFL). Traditional FL often suffers an accuracy drop during aggregation due to its averaging function, while SL and SFL face increased loss due to the independent gradient updates on each split network. TL adopts a unique strategy where the model traverses the nodes during forward propagation (FP) and performs backward propagation (BP) at the orchestrator, effectively implementing centralised learning (CL) principles within a distributed environment. The orchestrator is tasked with generating virtual batches and planning the model's sequential node visits during FP, aligning them with the ordered index of the data within these batches. We conducted experiments on six datasets representing diverse characteristics across various domains. Our evaluation demonstrates that TL is on par with classic CL approaches in terms of accurate inference, thereby offering a viable and robust solution for DL tasks. TL outperformed other DL methods and improved accuracy by 7.85% for independent and identically distributed (IID) datasets, macro F1‐score by 1.06% for non‐IID datasets, accuracy by 2.05% for text classification and AUC by 1.41% and 2.82% for medical and financial datasets, respectively. By effectively preserving data privacy while maintaining performance, TL represents a significant advancement in DL methodologies.

A novel approach to ultra-short-term multi-step wind power predictions based on encoder–decoder architecture in natural language processing

With the development of wireless networks and mobile devices, interest on indoor localization systems (ILSs) has increased. In particular, Wi-Fi-based ILSs are widely used because of the good prediction accuracy without additional hardware. However, as the prediction accuracy decreases in environments with natural noise, some studies were conducted to remove it. So far, two representative methods, i.e., the filtering-based method and deep learning-based method, have shown a significant effect in removing natural noise. However, the prediction accuracy of these methods severely decreased under artificial noise caused by adversaries. In this paper, we introduce a new media access control (MAC) spoofing attack scenario injecting artificial noise, where the prediction accuracy of Wi-Fi-based indoor localization system significantly decreases. We also propose a new deep learning-based indoor localization method using random forest(RF)-filter to provide the good prediction accuracy under the new MAC spoofing attack scenario. From the experimental results, we show that the proposed indoor localization method provides much higher prediction accuracy than the previous methods in environments with artificial noise.

Accuracy of genomic prediction within and across populations for nematode resistance and body weight traits in sheep

Polygenic risk scores (PRS) evaluate the individual genetic liability to a certain trait and are expected to play an increasingly important role in clinical risk stratification. Most often, PRS are estimated based on summary statistics of univariate effects derived from genome-wide association studies. To improve the predictive performance of PRS, it is desirable to fit multivariable models directly on the genetic data. Due to the large and high-dimensional data, a direct application of existing methods is often not feasible and new efficient algorithms are required to overcome the computational burden regarding efficiency and memory demands. We develop an adapted component-wise L2-boosting algorithm to fit genotype data from large cohort studies to continuous outcomes using linear base-learners for the genetic variants. Similar to the snpnet approach implementing lasso regression, the proposed snpboost approach iteratively works on smaller batches of variants. By restricting the set of possible base-learners in each boosting step to variants most correlated with the residuals from previous iterations, the computational efficiency can be substantially increased without losing prediction accuracy. Furthermore, for large-scale data based on various traits from the UK Biobank we show that our method yields competitive prediction accuracy and computational efficiency compared to the snpnet approach and further commonly used methods. Due to the modular structure of boosting, our framework can be further extended to construct PRS for different outcome data and effect types—we illustrate this for the prediction of binary traits.

The rear axle is prone to fatigue fracture due to the complicated load impact throughout the vehicle’s service procedure. Real-time suspension health monitoring during operation is an excellent strategy for avoiding such situations. This paper takes the automobile rear axle as the research object and uses the deep learning model to study the rear axle health monitoring technology. A damage prediction model combining multi-Gaussian fitting and long-term and short-term memory neural network is proposed for automobile rear axle health monitoring. The feature extraction method based on multi-Gaussian fitting can obtain comprehensive power spectral density information with fewer feature parameters, which is conducive to improve the calculation efficiency. The LSTM model can accurately identify the fatigue damage value of the rear axle from the extracted characteristic information and ensure the high precision of the damage prediction results. Several deep learning-based comparison models are used to demonstrate that the GAPSD LSTM model has good prediction accuracy and computation efficiency at the same time. It has been proven that the GAPSD LSTM model can efficiently and accurately forecast the fatigue damage of the rear axle by comparing the damage prediction results of all deep learning models.