Local Bandwidth Research Articles

PyGRF: An Improved Python Geographical Random Forest Model and Case Studies in Public Health and Natural Disasters

ABSTRACTGeographical random forest (GRF) is a recently developed and spatially explicit machine learning model. With the ability to provide more accurate predictions and local interpretations, GRF has already been used in many studies. The current GRF model, however, has limitations in its determination of the local model weight and bandwidth hyperparameters, potentially insufficient numbers of local training samples, and sometimes high local prediction errors. Also, implemented as an R package, GRF currently does not have a Python version which limits its adoption among machine learning practitioners who prefer Python. This work addresses these limitations by introducing theory‐informed hyperparameter determination, local training sample expansion, and spatially weighted local prediction. We also develop a Python‐based GRF model and package, PyGRF, to facilitate the use of the model. We evaluate the performance of PyGRF on an example dataset and further demonstrate its use in two case studies in public health and natural disasters.

Load Balanced PIM-Based Graph Processing

Graph processing is widely used for many modern applications, such as social networks, recommendation systems, and knowledge graphs. However, processing large-scale graphs on traditional Von Neumann architectures is challenging due to the irregular graph data and memory-bound graph algorithms. Processing-in-memory (PIM) architecture has emerged as a promising approach for accelerating graph processing by enabling computation to be performed directly on memory. Despite having many processing units and high local memory bandwidth, PIM often suffers from insufficient global communication bandwidth and high synchronization overhead due to load imbalance. This article proposes GraphB, a novel PIM-based graph processing system, to address all these issues. From the algorithm perspective, we propose a degree-aware graph partitioning algorithm that can generate balanced partitioning at a low cost. From the architecture perspective, we introduce tile buffers incorporated with an on-chip 2D-Mesh, which provides high bandwidth for inter-node data transfer. Dataflow in GraphB is designed to enable computation–communication overlap and dynamic load balancing. In a PyMTL3-based cycle-accurate simulator with five real-world graphs and three common algorithms, GraphB achieves an average 2.2× and maximum 2.8× speedup compared to the SOTA PIM-based graph processing system GraphQ.

Experimental Study on Rock Deformation Localization Using Digital Image Correlation and Acoustic Emission

In this study, the digital image correlation (DIC) method and acoustic emission (AE) technology were combined to study the evolution of rock deformation localization in detail. The second-order spatial–temporal subset DIC (STS-DIC) algorithm was proposed and used for measuring strongly heterogeneous deformation fields of red sandstone specimens under uniaxial compression. The evolution of the deformation field was analyzed with a focus on the deformation localization stage. The length and width of the deformation localization band (DLB) were measured, and the relationships between the relative sliding rate of the DLB, the relative opening rate of the DLB, and the AE counts were identified. Deformation localization was found to result from the rapid evolution of the strain concentration before the peak stress. The complete development of the DLB is an inducing factor for catastrophic rock failure, and the failure modes of the rock specimens were consistent with the final state of the DLB. A good correlation was identified between the AE counts and the relative displacement rate of the DLB, and the sliding rate was found to have a significant influence on the AE counts.

Improved estimation of drift coefficients using optimal local bandwidths

Stochastic differential equations (SDEs) are commonly used to model various systems. Data-driven methods have been widely used to estimate the drift and diffusion terms of a Langevin equation. Among the most commonly used estimation methods is the Nadaraya–Watson estimator, which is a non-parametric data-driven approach. In this study, we propose a method to improve the estimation of the drift coefficient of a stochastic process using optimal local bandwidths that minimize the error of the approximation of the first conditional moments of a univariate system. This approach is compared to a global bandwidth estimation and an estimation based on a fixed number of nearest neighbors. The proposed method has the potential to reduce the error of the drift estimation, thereby improving the accuracy of the model.

Spectral subspaces of Sturm-Liouville operators and variable bandwidth

We study spectral subspaces of the Sturm-Liouville operator f↦−(pf′)′ on R, where p is a positive, piecewise constant function. Functions in these subspaces can be thought of as having a local bandwidth determined by 1/p. Using the spectral theory of Sturm-Liouville operators, we make the reproducing kernel of these spectral subspaces more explicit and compute it completely in certain cases. As a contribution to sampling theory, we then prove necessary density conditions for sampling and interpolation in these subspaces and determine the critical density that separates sets of stable sampling from sets of interpolation.

Energy attenuation of seismic metamaterials composed of a periodic array of coated elliptical cylinders

Abstract We present a numerical study on energy attenuation of seismic metamaterials consisting of a periodic array of coated elliptical cylinders. The aim is to perceive the effect of aspect ratio for different wave modes so that the metamaterials can interact with the incoming wave causing them to interfere with each other destructively, especially for low-frequency seismic waves with relatively wide bandgap. Previous studies mainly focused on the configuration of coated circular cylinders or spheres, in which the metamaterial is composed of a hard inclusion surrounded by a soft coating layer and dispersed within a hard matrix. Here we utilize numerical simulations based on finite element calculation to analyze the local fields within the unit cell. Effective mass density, mass moment of inertia and shear modulus are analyzed through a homogenization procedure to characterize the macroscopic behavior of the effective medium. The effective behavior will be dependent for different aspect ratios and for different types of wave motions. To verify the effectiveness of energy attenuation, a full-scale model is adopted. Specifically, to identify optimal energy attenuation configurations, we illustrate the attenuation effects of elliptical metamaterials under longitudinal and shear horizontal types of waves. The present study demonstrates that elliptical metamaterials will have more reflexibilities to tune with the aspect ratio of the elliptical geometry as well as the directionality of incidence waves. Based on our simulations, we show the ability of the designed configuration in tuning local resonance frequencies and bandwidths for real implementations and applications of seismic metamaterials.

On the weak convergence and the uniform-in-bandwidth consistency of the general conditional $U$-processes based on the copula representation: multivariate setting

$U$-statistics represent a fundamental class of statistics from modeling quantities of interest defined by multi-subject responses. $U$-statistics generalise the empirical mean of a random variable $X$ to sums over every $m$-tuple of distinct observations of $X$. Stute [Conditional U -statistics, Ann. Probab., 1991] introduced a class of estimators called conditional $U$-statistics. In the present work, we provide a new class of estimators of conditional $U$-statistics. More precisely, we investigate the conditional $U$-statistics based on copula representation. We establish the uniform-in-bandwidth consistency for the proposed estimator. In addition, uniform consistency is also established over $\varphi \in \mathscr{F}$ for a suitably restricted class $\mathscr{F}$, in both cases bounded and unbounded, satisfying some moment conditions. Our theorems allow data-driven local bandwidths for these statistics. Moreover, in the same context, we show the uniform bandwidth consistency for the nonparametric Inverse Probability of Censoring Weighted estimators of the regression function under random censorship, which is of its own interest. We also consider the weak convergence of the conditional $U$-statistics processes. We discuss the wild bootstrap of the conditional $U$-statistics processes. These results are proved under some standard structural conditions on the Vapnik-Chervonenkis class of functions and some mild conditions on the model.

Exploiting Cloud Object Storage for High-Performance Analytics

Elasticity of compute and storage is crucial for analytical cloud database systems. All cloud vendors provide disaggregated object stores, which can be used as storage backend for analytical query engines. Until recently, local storage was unavoidable to process large tables efficiently due to the bandwidth limitations of the network infrastructure in public clouds. However, the gap between remote network and local NVMe bandwidth is closing, making cloud storage more attractive. This paper presents a blueprint for performing efficient analytics directly on cloud object stores. We derive cost- and performance-optimal retrieval configurations for cloud object stores with the first in-depth study of this foundational service in the context of analytical query processing. For achieving high retrieval performance, we present AnyBlob , a novel download manager for query engines that optimizes throughput while minimizing CPU usage. We discuss the integration of high-performance data retrieval in query engines and demonstrate it by incorporating AnyBlob in our database system Umbra. Our experiments show that even without caching, Umbra with integrated AnyBlob achieves similar performance to state-of-the-art cloud data warehouses that cache data on local SSDs while improving resource elasticity.

Effect of the local energy distribution of x-ray beams generated through inverse Compton scattering in dual-energy imaging applications.

X-ray sources based on the inverse Compton interaction between a laser and a relativistic electron beam are emerging as a promising compact alternative to synchrotron for the production of intense monochromatic and tunable radiation. The emission characteristics enable several innovative imaging techniques, including dual-energy K-edge subtraction (KES) imaging. The performance of these techniques is optimal in the case of perfectly monochromatic x-ray beams, and the implementation of KES was proven to be very effective with synchrotron radiation. Nonetheless, the features of inverse Compton scattering (ICS) sources make them good candidates for a more compact implementation of KES techniques. The energy and intensity distribution of the emitted radiation is related to the emission direction, which means different beam qualities in different spatial positions. In fact, as the polar angle increases, the average energy decreases, while the local energy bandwidth increases and the emission intensity decreases. The scope of this work is to describe the impact of the local energy distribution variations on KES imaging performance. By means of analytical simulations, the reconstructed signal, signal-to-noise ratio, and background contamination were evaluated as a function of the position of each detector pixel. The results show that KES imaging is possible with ICS x-ray beams, even if the image quality slightly degrades at the detector borders for a fixed collimation angle and, in general, as the beam divergence increases. Finally, an approach for the optimization of specific imaging tasks is proposed by considering the characteristics of a given source.

Tuning the Mottness in Sr3Ir2O7 via Bridging Oxygen Vacancies

The electronic evolution of Mott insulators into exotic correlated phases remains puzzling, because of electron interaction and inhomogeneity. Introduction of individual imperfections in Mott insulators could help capture the main mechanism and serve as a basis to understand the evolution. Here we utilize scanning tunneling microscopy to probe the atomic scale electronic structure of the spin-orbit-coupling assisted Mott insulator Sr3Ir2O7. It is found that the tunneling spectra exhibit a homogeneous Mott gap in defect-free regions, but near the oxygen vacancy in the rotated IrO2 plane the local Mott gap size is significantly enhanced. We attribute the enhanced gap to the locally reduced hopping integral between the 5d electrons of neighboring Ir sites via the bridging planar oxygen p orbitals. Such bridging defects have a dramatic influence on local bandwidth, thus provide a new way to manipulate the strength of Mottness in a Mott insulator.

A metamaterial plate with magnetorheological elastomers and gradient resonators for tuneable, low-frequency and broadband flexural wave manipulation

Incorporating magnetorheological elastomers and multiple gradient resonators (MRE–MGRs), a novel metamaterial plate is proposed, which entails low-frequency, broadband and tuneable bandgap features. The proposed MRE-MGR design embeds an adaptive magnetic field adjustment mechanism to alter the elastic modulus of the MRE to modulate the local resonance bandgap frequency and bandwidth. A bandgap prediction model of the metamaterial plate is developed using the plane wave expansion (PWE) method. Integrated into the model through equivalent stiffness terms, the magneto-controlled modulus of the MRE can be estimated using the energy method and magnetic dipole theory. The established model allows for revealing the dispersion relation of the proposed metamaterial plate under various magnetic fields. The predicted tuneable bandgaps, vibration transmission loss and flexural wave modulation of the metamaterial plate are validated through comparisons with finite element simulations. Numerical analyses show the benefit of the gradient resonator design, which alongside the effective MRE tuning, entails broadening low frequency bandgaps at 92 Hz (relative band of 79.3% for the exponential gradient) and 72.7 Hz (relative band of 65.9% for the linear gradient). The mechanisms underpinning the observed bandgap widening phenomenon are analysed and attributed to the combined effects arising from the gradient design of the distributed resonators and the effective tuning of the MRE elastic modulus under magnetic field modulation. The proposed metamaterial plate functionally achieves simultaneous magnetron adjustment of the bandgap frequency and its width, thus holding great promise for applications such as flexural wave guiding as well as vibration and sound radiation control in planar structures.

High Efficiency Vibrotactile Codec Based on Gate Recurrent Network

The multimedia has achieved dominant positions in both local storage and internet bandwidth, which inevitably promotes the compression of audio, image and video information. Nowadays, the emerging haptic technology, which enhances the immersion in virtual reality and remote control, has also brought new challenges in its codec design. It is thus imperative to develop haptic codecs, including kinesthetic and vibrotactile codecs, with high efficiency and low delay. In this paper, we exploit statistical features of vibrotactile data to develop a Recurrent-Network-based Vibrotactile Codec (RNVC) with high compression efficiency and low coding delay. The proposed encoder consists of vibrotactile estimation by Gate Recurrent Unit (GRU), non-uniform quantization/compensation of residuals and an entropy encoder. In particular, the GRU-based recurrent network is utilized for its high efficiency to predict signals and low complexity to converge. The decoder consists of all counterparts of encoder. Experimental results show the proposed RNVC significantly reduces of original bitrates with negligible encoding delay, which achieves the state-of-the-art coding performance of vibrotactile signal.

Microwave Contactless Current-Sensing for Live/Dead Differentiation of Single Bioparticles on a Microfluidic Platform

The combination of microwave and microfluidic technologies enables the creation of efficient platforms to rapidly quantify live/dead bioparticles in industrial processes, biomedical research and environmental monitoring applications. In this regard, this work demonstrates the differentiation of single live/dead bioparticles with a microwave-microfluidic platform. The system is composed of a contactless current-sensing superheterodyne architecture with broadband frequency-operation bowtie electrodes. Furthermore, the 20mm × 20mm bowtie electrode has been designed and manufactured with a –10 dB frequency range, from 4GHz to 8GHz, and capable of confining the electric field in a micrometric volume. The gap of the bowtie geometry of the electrodes is adjusted to the dimensions of the 50 μm × 50 μm cross-section and 40mm long microfluidic channel. In particular, 10mL sample with 10 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">9</sup> bacteria/mL flows through the microchannel at 19.8 μL/min, focused with elasto-inertial effect enhanced with 500 ppm of polyethylene oxide. Accordingly, from the signal captured by the electrodes, the energy variation of the applied electric field produced by the bioparticles is sensed by a current-based transimpedance amplifier. Here, the incorporation of the transimpedance amplifier in the measurement system improves the SNR by 4 dB. In addition, the superheterodyne transceiver has been optimized in terms of local oscillator power and bandwidth. Furthermore, the microfluidic system has been adjusted so that the balance of hydrodynamic forces confines bioparticles at the centerline of the microchannel within less than 5 μm. As a result, the system measures the cytoplasmic content of live and dead flowing bacteria with a difference of 6 dB and relative deviation of 3.2%.

A cloud-based consultation and collaboration system for radiotherapy: Remote decision support services for community radiotherapy centers

PurposeThis study aimed to establish a cloud-based radiotherapy consultation and collaboration system, then investigated the practicability of remote decision support for community radiotherapy centers using the system. Methods and MaterialsA cloud-based consultation and collaboration system for radiotherapy, OncoEvidance®, was developed to provide remote services of LINAC modeling, simulation CT data import/export, target volume and organ-at-risk delineation, prescription, and treatment planning. The system was deployed on a hybrid cloud. A federate of public nodes, each corresponding to a medical institution, are managed by a central node where a group of consultants have registered. Users can access the system through network using computing devices. The system has been tested at three community radiotherapy centers. One accelerator was modeled. 12 consultants participated the remote radiotherapy decision support and 77 radiation treatment plans had been evaluated remotely. ResultsAll the passing rates of per-beam dose verification are > 94% and all the passing rates of composite beam dose verification are > 99%. The average downloading time for one set of simulation CT data for one patient from Internet was within 1 min under the cloud download bandwidth of 8 Mbps and local network bandwidth of 100 Mbps. The average response time for one consultant to contour target volumes and make prescription was about 24 h. And that for one consultant to design and optimize a IMRT treatment plan was about 36 h. 100% of the remote plans passed the dosimetric criteria and could be imported into the local TPS for further verification. ConclusionThe cloud-based consultation and collaboration system saved the travel time for consultants and provided high quality radiotherapy to patients in community centers. The under-staffed community radiotherapy centers could benefit from the remote system with lower cost and better treatment quality control.

Analysis method for calculating radial correlation length of electron temperature turbulence from correlation electron cyclotron emission radiometer.

The radial correlation length (Lr) is one of the essential quantities to measure in order to more fully characterize and understand turbulence and anomalous transport in magnetic fusion plasmas. The analysis method for calculating Lr of electron temperature (Te) turbulence from correlation electron cyclotron emission (correlation ECE or CECE) radiometer measurements has not been fully developed partly due to the fact that the turbulent electron temperature fluctuations are generally imbedded in much larger amplitude thermal noise, which leads to a greatly reduced cross correlation coefficient (ϱ) between two spatially separated ECE signals. This work finds that this ϱ reduction factor due to thermal noise is a function of the local relative temperature fluctuation power and CECE system bandwidths of intermediate and video frequencies, independent of radial separations. This indicates that under the approximation of constant relative temperature fluctuation power for a small radial range of local CECE measurements, the original shape of ϱ as a function of radial separation without thermal noise is preserved in the CECE data with thermal noise present. For Te turbulence with a Gaussian radial wavenumber spectrum, a fit function using the product of Gaussian and sinusoidal functions is derived for calculating Lr. This analysis method has been numerically tested using simulated ECE radiometer data over a range of parameters. Using this method, the experimental temperature turbulence correlation length Lr in a DIII-D L-mode plasma is found to be ∼10 times the local ion gyroradius.

Variational multi-harmonic mode extraction for characterising impulse envelope of bearing failures

The envelope shape of a failure-induced impulse response reflects the strike procedure of the bearing failure area and provides information on the size and contour of the defective area. Because the impulse envelope is a broad-bandwidth component severely affected by in-band noise, it is difficult to separate from the vibration signals. To eliminate the in-band noise and obtain diagnostic information from the full-band envelope, a novel method, called variational multi-harmonic mode extraction (VMHME), is proposed to extract the impulse envelope component and characterise the envelope shape. First, a multi-harmonic mode function (MHMF) is constructed to define a harmonic assembly, which is generally has a broad-bandwidth property, but each harmonic is narrow-bandwidth. A variational model is then established to optimally decompose the mode with explicit MHMF from the analysed signal. Only one specific mode is extracted at a time and adopted to efficiently satisfy the demand for failure-induced impulse envelope extraction. Essentially, VMHME provides an optimal band-pass filter bank just with a local narrow bandwidth at failure characteristic harmonics. Utilising​ the narrow-bandwidth harmonic assembly eliminates the in-band noise, whereas the failure characteristics in the envelope shape are retained The effectiveness of the proposed method was verified using both simulated and experimental bearing failure signals. The results prove that VMHME can make in-depth use of the envelope shape information for bearing failure diagnosis.

Joint Optimization of Resource Allocation and UAVs’ Trajectories in Multi-UAV-Aided MEC Systems

The combination of unmanned aerial vehicles (UAVs) and mobile edge computing (MEC) has become an up-and-coming technique to provide additional computing support for smart mobile devices with limited battery capacity in the wireless communication network systems. In this paper, a multi-UAV-aided MEC system is modelled where UAVs are equipped with MEC servers and users can execute local computing and computation offloading to UAVs simultaneously. The total amount of processing bits of all users is maximized by the joint optimization of the user-UAV association, local processing frequency, transmit power, bandwidth allocation and UAVs’ trajectories, meanwhile, the energy consumption constraint and the UAVs’ speed constraint are satisfied. Because the formulated problem is not convex, it is challenging to get the global optimal solution directly. To iron out the non-convex problem, it is divided into three subproblems and solved by an alternating optimization algorithm. Simulations show that the algorithm introduced in our paper performs better than the local computing, full offloading and fixed trajectory patterns.

Adaptive manifold density estimation

When data concentrate on lower dimensional manifolds embedded in a higher dimensional representation space and these manifolds have different intrinsic geometry and dimensions, density estimators with locally adaptive smoothing parameters show substantial gains over those with fixed bandwidths. However, it is difficult to decide how the local bandwidth parameters should be parameterized. In this paper, we propose a sample-point density estimator. For each observation, we associate it with a unique bandwidth parameter which is achieved by computing the local covariance matrix of its k nearest neighbours. Compared to the fixed choice of the neighbourhood size for all the samples in [Vincent P, Bengio Y. Manifold parzen windows. In: Becker S, Thrun S, and Obermayer K, editors, Advances in neural information processing systems Vol. 15. MIT Press; 2003. p. 849–856], our adaptive manifold density estimator allows samples to select their own neighbourhood sizes.

Nonparametric quantile regression with missing data using local estimating equations

In this paper, we propose augmented inverse probability weighted (AIPW) local estimating equations in dealing with missing data in nonparametric quantile regression context. The missing mechanism here is missing at random. To avoid the problem of misspecification, we adopt nonparametric approach to estimate the propensity score and conditional expectations of estimating functions. The asymptotic properties of our proposed estimator are studied. Majorisation–minimisation algorithm is used to circumvent the nonsmoothness of check function at the origin. When it comes to the choice of bandwidth, the theoretical expression of local optimal bandwidth is derived based on asymptotic properties. Moreover, we apply smoothed bootstrap method to obtain the empirical mean square error and use cross-validation to determine the bandwidth in practice. Simulations are conducted to compare the performance of our proposed methods with other existing methods. Finally, we illustrate our methodology with an analysis of non-insulin-dependent diabetes mellitus data set.

Spatial Data Dependence Graph Based Pre-RTL Simulator for Convolutional Neural Network Dataflows

In this paper, a new pre-RTL simulator is proposed to predict the power, performance, and area of convolutional neural network (CNN) dataflows prior to register-transfer-level (RTL) design. In the simulator, a novel approach is adopted to implement a spatial data dependence graph (SDDG), which enables us to model a specific dataflow alongside inter-instruction dependencies by tracking the status of each processing element (PE). In addition, the proposed pre-RTL simulator makes it possible to evaluate the impact of memory constraints such as latency and bandwidth. The latency-insensitive and bandwidth-insensitive PE controllers assumed in the proposed pre-RTL simulator guarantee both functional correctness and maximum performance, regardless of memory constraints. In particular, it is shown that the optimal distribution method of local memory bandwidth can reduce the accelerator execution time by up to 37.6% compared with the equal distribution method. For weight stationary (WS) and row stationary (RS) dataflows, the accelerator performance closely depends on memory constraints. The simulation results also show that the relative performances of dataflows depend on the layer shape of the convolutional layer. For example, for an identical hardware area in a standard convolutional layer of AlexNet, WS dataflows do not provide any performance gain over RS dataflows when the memory latency is sufficiently high. In addition, WS dataflows cannot fully reuse the input activation, thereby increasing local memory accesses, since the number of weights loaded at a specific time is limited. Moreover, in a depth-wise convolutional layer of MobileNet, WS dataflows tend to outperform RS dataflows even in the presence of large memory latency. The source code is available on the GitHub repository: <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/SDL-KU/SDDGSim</uri> .