Prevalent large CNN models pose a significant challenge in terms of computing resources for resource-constrained devices at the Edge. Distributing the computations and coefficients over multiple edge devices collaboratively has been well studied but these works generally do not consider the presence of device failures (e.g., due to temporary connectivity issues, overload, discharged battery of edge devices). Such unpredictable failures can compromise the reliability of edge devices, inhibiting the proper execution of distributed CNN inference. In this paper, we present a novel partitioning method, called RobustDiCE, for robust distribution and inference of CNN models over multiple edge devices. Our method can tolerate intermittent and permanent device failures in a distributed system at the Edge, offering a tunable trade-off between robustness (i.e., retaining model accuracy after failures) and resource utilization . We verify the system’s robustness by validating the overall end-to-end latency under failures. We evaluate RobustDiCE using the ImageNet-1K dataset on several representative CNN models under various device failure scenarios and compare it with several state-of-the-art partitioning methods as well as an optimal robustness approach (i.e., full neuron replication). In addition, we demonstrate RobustDiCE’s advantages in terms of memory usage and energy consumption per device, and system throughput for various system setups with different device counts.

Prevalent large CNN models pose a significant challenge in terms of computing resources for resource-constrained devices at the Edge. Distributing the computations and coefficients over multiple edge devices collaboratively has been well studied but these works generally do not consider the presence of device failures (e.g., due to temporary connectivity issues, overload, discharged battery, etc. of edge devices). Such unpredictable failures can compromise the reliability of edge devices, inhibiting the proper execution of distributed CNN inference. In this paper, we present a novel partitioning method, called RobustDiCE, for robust distribution and inference of CNN models over multiple edge devices. Our method can tolerate intermittent and permanent device failures in a distributed system at the Edge, offering a tunable trade-off between robustness (i.e., retaining model accuracy after failures) and resource utilization. We evaluate RobustDiCE using the ImageNet-1K dataset on several representative CNN models under various device failure scenarios and compare it with several state-of-the-art partitioning methods as well as an optimal robustness approach (i.e., full neuron replication). In addition, we demonstrate RobustDiCE’s advantages in terms of memory usage and energy consumption per device, and system throughput for various system set-ups with different device counts.

Prevalent large transformer models present significant computational challenges for resource-constrained devices at the Edge. While distributing the workload of deep learning models across multiple edge devices has been extensively studied, these works typically overlook the impact of failures of edge devices. Unpredictable failures, due to, e.g., connectivity issues or discharged batteries, can compromise the reliability of inference serving at the Edge. In this article, we introduce a novel methodology, called EASTER, designed to learn robust distribution strategies for transformer models against device failures that consider the tradeoff between robustness (i.e., maintaining model functionality against failures) and resource utilization (considering memory usage and computations). We evaluate EASTER with three representative transformers—ViT, GPT-2, and Vicuna—under device failures. Our results demonstrate EASTER’s efficiency in memory usage, and possible end-to-end latency improvement for inference across multiple edge devices while preserving model accuracy as much as possible under device failures.

Abstract Because involving a large number of discrete particles interacting with water, simulating the fluid–solid coupling problem of landslide‐generated waves in a virtual environment is very difficult. This paper employs the DualSPHysics multiphase flow model to relatively easily simulate the dynamics of landslide‐generated waves in large scales. In addition, the simulation results are well visualized by the marching cubes algorithm. It is found that, when the factors of the landslide scale, the water length, and water depth are fixed, the initial maximum wave height decreases as the water width increases. After the water is squeezed by the falling slider, one‐way and two‐way propagations of the surge are generated under 2D and 3D conditions, respectively. In addition, under the true‐3D condition, although the initial swell generated is smaller than the 2D and quasi‐2D conditions, it will generate a large wave height during the climbing and undulation of the near shore. In the 2D condition, the surge generated is almost the largest. Therefore, when using the DualSPHysics multiphase flow model to predict the maximum wave height generated by landslide surges, a 2D simulation can be used to improve work efficiency.

Recently, the acquisition of high-resolution T1maps in a clinically feasible time frame has been demonstrated with Driven Equilibrium Single Pulse Observation of T1(DESPOT1). DESPOT1 derives the longitudinal relaxation time, T1, from two or more spoiled gradient recalled echo (SPGR) images acquired with a constant TRand different flip angles. In general, T1can be estimated from two or more SPGR images. Estimation of MR parameters (T1, M0, etc.) from these sequences is challenging and susceptible to the level of noise in signal acquisition. Methods such as Simplex Optimization, Weighted Non-Linear Least Squares (WNLS), Linear Least Square (LLS or Gupta's LLS), and Intensity based Linear Least Square (ILLS) method have been employed to estimate T1. In both linear and non-linear methods, the estimated T1values are highly dependent on defining the weighting factors; errors in these weighting factors can result in a biased estimate of T1. In this study, an adaptive neural network (ANN) is introduced, trained and evaluated. The ANN was trained using an analytical model of the SPGR signal in the presence of different levels of signal to noise ratio (2 to 30). Receiver Operator Characteristic (ROC) analysis and the K-fold cross-validation (KFCV) method were employed to train, test, and optimize the network. The result (Az=0.81) shows that, compared to the other techniques, ANNs can provide a faster and unbiased estimate of T1from SPGR signals.

Magnetic Resonance Imaging (MRI) estimation of contrast agent concentration in fast pulse sequences such as Dual Gradient Echo (DGE) imaging is challenging. An Adaptive Neural Network (ANN) was trained with a map of contrast agent concentration estimated by Look-Locker (LL) technique (modified version of inversion recovery imaging) as a gold standard. Using a set of features extracted from DGE MRI data, an ANN was trained to create a voxel based estimator of the time trace of CA concentration. The ANN was trained and tested with the DGE and LL information of six Fisher rats using a K-Fold Cross-Validation (KFCV) method with 60 folds and 10500 samples. The Area Under the Receiver Operator Characteristic Curve (AUROC) for 60 folds was used for training, testing and optimization of the ANN. After training and optimization, the optimal ANN (4:7:5:1) produced maps of CA concentration which were highly correlated (r = 0.89, P <; 0.0001) with the CA concentration estimated by the LL technique. The estimation made by the ANN had an excellent overall performance (AUROC = 0.870).

Diffusion kurtosis imaging (DKI) is a recent MRI based method that can quantify deviation from Gaussian behavior using a kurtosis tensor. DKI has potential value for the assessment of neurologic diseases. Existing techniques for diffusion kurtosis imaging typically need to capture hundreds of MRI images, which is not clinically feasible on human subjects. In this paper, we develop robust denoising and model fitting methods that make it possible to accurately reconstruct a kurtosis tensor from 75 or less noisy measurements. Our denoising method is based on subspace learning for multi-dimensional signals and our model fitting technique uses iterative reweighting to effectively discount the influences of outliers. The total data acquisition time thus drops significantly, making diffusion kurtosis imaging feasible for many clinical applications involving human subjects.

Magnetic susceptibility is an important physical property of tissues, and can be used as a contrast mechanism in magnetic resonance imaging (MRI). Recently, targeting contrast agents by conjugation with signaling molecules and labeling stem cells with contrast agents have become feasible. These contrast agents are strongly paramagnetic, and the ability to quantify magnetic susceptibility could allow accurate measurement of signaling and cell localization. Presented here is a technique to estimate arbitrary magnetic susceptibility distributions by solving an ill-posed inversion problem from field maps obtained in an MRI scanner. Two regularization strategies are considered: conventional Tikhonov regularization and a sparsity promoting nonlinear regularization using the l(1) norm. Proof of concept is demonstrated using numerical simulations, phantoms, and in a stroke model in a rat. Initial experience indicates that the nonlinear regularization better suppresses noise and streaking artifacts common in susceptibility estimation.

Fast and accurate measurement of the longitudinal relaxation time, T1, has become increasingly important in quantitative estimates of such tissue physiological parameters as perfusion, capillary permeability, and tissue interstitial space using dynamic contrast-enhanced MRI (DCE-MRI). The look-locker (LL) sequence provides accurate T1estimates, with the advantages of reduced acquisition time, and a wide range of sampling times post-inversion. In this study, an adaptive neural network (ANN) was trained and employed as an unbiased estimator of T1. The ANN estimator was trained by simulating the LL signal at different levels of SNR. The results of its application to the simulated data were compared with T1maps estimated by conventional methods (simplex method with non-negative least-squares fitting). Experimental results of the ANN method for 19 animals were also compared to the the conventional method, and with values of T1reported in literature. The ANN and conventional methods produce estimates that are highly correlated in normal (r = 0.957, p1map in tissue, and thus to estimate from LL data in DCE studies the temporal change in tissue R1that occurs after administration of contrast agent, a measure that plays an important role in quantification of physiological parameters using MRI.