scholarly journals Multi-core CPU or GPU-accelerated Multiscale Modeling for Biomolecular Complexes

2013 ◽  
Vol 1 ◽  
pp. 164-179 ◽  
Author(s):  
Tao Liao ◽  
Yongjie Zhang ◽  
Peter M. Kekenes-Huskey ◽  
Yuhui Cheng ◽  
Anushka Michailova ◽  
...  
Keyword(s):  
Graphics Processing Units ◽  
Grid Point ◽  
Data Bank ◽  
Kernel Functions ◽  
Gaussian Kernel ◽  
Multi Scale ◽  
Biomolecular Complexes ◽  
Biomolecular Surface ◽  
Grid Points ◽  
Structure Indices

AbstractMulti-scale modeling plays an important role in understanding the structure and biological functionalities of large biomolecular complexes. In this paper, we present an efficient computational framework to construct multi-scale models from atomic resolution data in the Protein Data Bank (PDB), which is accelerated by multi-core CPU and programmable Graphics Processing Units (GPU). A multi-level summation of Gaussian kernel functions is employed to generate implicit models for biomolecules. The coefficients in the summation are designed as functions of the structure indices, which specify the structures at a certain level and enable a local resolution control on the biomolecular surface. A method called neighboring search is adopted to locate the grid points close to the expected biomolecular surface, and reduce the number of grids to be analyzed. For a specific grid point, a KD-tree or bounding volume hierarchy is applied to search for the atoms contributing to its density computation, and faraway atoms are ignored due to the decay of Gaussian kernel functions. In addition to density map construction, three modes are also employed and compared during mesh generation and quality improvement to generate high quality tetrahedral meshes: CPU sequential, multi-core CPU parallel and GPU parallel. We have applied our algorithm to several large proteins and obtained good results.

scholarly journals Image-based textile decoding

2020 ◽  
pp. 1-14
Author(s):  
Siqiang Chen ◽  
Masahiro Toyoura ◽  
Takamasa Terada ◽  
Xiaoyang Mao ◽  
Gang Xu
Keyword(s):  
Deep Learning ◽  
Grid Point ◽  
Intermediate Representation ◽  
Training Set ◽  
Binary Matrix ◽  
The Matrix ◽  
Correct Pattern ◽  
Grid Points ◽  
Textile Fabric ◽  
Direct Correspondence

A textile fabric consists of countless parallel vertical yarns (warps) and horizontal yarns (wefts). While common looms can weave repetitive patterns, Jacquard looms can weave the patterns without repetition restrictions. A pattern in which the warps and wefts cross on a grid is defined in a binary matrix. The binary matrix can define which warp and weft is on top at each grid point of the Jacquard fabric. The process can be regarded as encoding from pattern to textile. In this work, we propose a decoding method that generates a binary pattern from a textile fabric that has been already woven. We could not use a deep neural network to learn the process based solely on the training set of patterns and observed fabric images. The crossing points in the observed image were not completely located on the grid points, so it was difficult to take a direct correspondence between the fabric images and the pattern represented by the matrix in the framework of deep learning. Therefore, we propose a method that can apply the framework of deep learning viau the intermediate representation of patterns and images. We show how to convert a pattern into an intermediate representation and how to reconvert the output into a pattern and confirm its effectiveness. In this experiment, we confirmed that 93% of correct pattern was obtained by decoding the pattern from the actual fabric images and weaving them again.

scholarly journals A Fast and Lightweight Method with Feature Fusion and Multi-Context for Face Detection

2018 ◽  
Vol 10 (8) ◽  
pp. 80
Author(s):  
Lei Zhang ◽  
Xiaoli Zhi
Keyword(s):  
Face Detection ◽  
Graphics Processing Units ◽  
High Performance ◽  
Feature Fusion ◽  
Local Context ◽  
Data Set ◽  
Global Context ◽  
Detection Algorithms ◽  
Multi Scale ◽  
Benchmark Datasets

Convolutional neural networks (CNN for short) have made great progress in face detection. They mostly take computation intensive networks as the backbone in order to obtain high precision, and they cannot get a good detection speed without the support of high-performance GPUs (Graphics Processing Units). This limits CNN-based face detection algorithms in real applications, especially in some speed dependent ones. To alleviate this problem, we propose a lightweight face detector in this paper, which takes a fast residual network as backbone. Our method can run fast even on cheap and ordinary GPUs. To guarantee its detection precision, multi-scale features and multi-context are fully exploited in efficient ways. Specifically, feature fusion is used to obtain semantic strongly multi-scale features firstly. Then multi-context including both local and global context is added to these multi-scale features without extra computational burden. The local context is added through a depthwise separable convolution based approach, and the global context by a simple global average pooling way. Experimental results show that our method can run at about 110 fps on VGA (Video Graphics Array)-resolution images, while still maintaining competitive precision on WIDER FACE and FDDB (Face Detection Data Set and Benchmark) datasets as compared with its state-of-the-art counterparts.

Optimization of the Kernel Functions in a Probabilistic Neural Network Analyzing the Local Pattern Distribution

2002 ◽  
Vol 14 (5) ◽  
pp. 1183-1194 ◽  
Author(s):  
I. Galleske ◽  
J. Castellanos
Keyword(s):  
Neural Network ◽  
Covariance Matrix ◽  
Probabilistic Neural Network ◽  
Local Environment ◽  
Kernel Functions ◽  
Gaussian Kernel ◽  
Classification Rate ◽  
Training Pattern ◽  
The Neural Network ◽  
Pattern Distribution

This article proposes a procedure for the automatic determination of the elements of the covariance matrix of the gaussian kernel function of probabilistic neural networks. Two matrices, a rotation matrix and a matrix of variances, can be calculated by analyzing the local environment of each training pattern. The combination of them will form the covariance matrix of each training pattern. This automation has two advantages: First, it will free the neural network designer from indicating the complete covariance matrix, and second, it will result in a network with better generalization ability than the original model. A variation of the famous two-spiral problem and real-world examples from the UCI Machine Learning Repository will show a classification rate not only better than the original probabilistic neural network but also that this model can outperform other well-known classification techniques.

Spatio-temporal variability of temperature and potential evapotranspiration over India

2016 ◽  
Vol 7 (4) ◽  
pp. 810-822 ◽  
Author(s):  
P. Sonali ◽  
D. Nagesh Kumar
Keyword(s):  
Hydrological Cycle ◽  
Potential Evapotranspiration ◽  
Grid Point ◽  
Rank Correlation ◽  
Trend Detection ◽  
Detection Analysis ◽  
Change Impact ◽  
Spatio Temporal ◽  
Grid Points ◽  
Significant Factors

Worldwide, major changes in the climate are expected due to global warming, which leads to temperature variations. To assess the climate change impact on the hydrological cycle, a spatio-temporal change detection study of potential evapotranspiration (PET) along with maximum and minimum temperatures (Tmax and Tmin) over India have been performed for the second half of the 20th century (1950–2005) both at monthly and seasonal scale. From the observed monthly climatology of PET over India, high values of PET are envisioned during the months of March, April, May and June. Temperature is one of the significant factors in explaining changes in PET. Hence seasonal correlations of PET with Tmax and Tmin were analyzed using Spearman rank correlation. Correlation of PET with Tmax was found to be higher compared to that with Tmin. Seasonal variability of trend at each grid point over India was studied for Tmax, Tmin and PET separately. Trend Free Pre-Whitening and Modified Mann Kendall approaches, which consider the effect of serial correlation, were employed for the trend detection analysis. A significant trend was observed in Tmin compared to Tmax and PET. Significant upward trends in Tmax, Tmin and PET were observed over most of the grid points in the interior peninsular region.

Exact Subthreshold Integration with Continuous Spike Times in Discrete-Time Neural Network Simulations

2007 ◽  
Vol 19 (1) ◽  
pp. 47-79 ◽  
Author(s):  
Abigail Morrison ◽  
Sirko Straube ◽  
Hans Ekkehard Plesser ◽  
Markus Diesmann
Keyword(s):  
Grid Point ◽  
Propagation Delay ◽  
Integration Scheme ◽  
The Novel ◽  
Neuron Models ◽  
Exact Integration ◽  
Wide Range ◽  
Network Simulations ◽  
Adequate Representation ◽  
Grid Points

Very large networks of spiking neurons can be simulated efficiently in parallel under the constraint that spike times are bound to an equidistant time grid. Within this scheme, the subthreshold dynamics of a wide class of integrate-and-fire-type neuron models can be integrated exactly from one grid point to the next. However, the loss in accuracy caused by restricting spike times to the grid can have undesirable consequences, which has led to interest in interpolating spike times between the grid points to retrieve an adequate representation of network dynamics. We demonstrate that the exact integration scheme can be combined naturally with off-grid spike events found by interpolation. We show that by exploiting the existence of a minimal synaptic propagation delay, the need for a central event queue is removed, so that the precision of event-driven simulation on the level of single neurons is combined with the efficiency of time-driven global scheduling. Further, for neuron models with linear subthreshold dynamics, even local event queuing can be avoided, resulting in much greater efficiency on the single-neuron level. These ideas are exemplified by two implementations of a widely used neuron model. We present a measure for the efficiency of network simulations in terms of their integration error and show that for a wide range of input spike rates, the novel techniques we present are both more accurate and faster than standard techniques.

scholarly journals Derivation of a planetary ionospheric storm index

2008 ◽  
Vol 26 (9) ◽  
pp. 2645-2648 ◽  
Author(s):  
T. L. Gulyaeva ◽  
I. Stanislawska
Keyword(s):  
Solar Cycle ◽  
Extreme Values ◽  
Grid Point ◽  
Radiant Energy ◽  
Electron Content ◽  
Vertical Total Electron Content ◽  
Ionospheric Storm ◽  
Total Electron ◽  
Grid Points ◽  
Storm Index

Abstract. The planetary ionospheric storm index, Wp, is deduced from the numerical global ionospheric GPS-IONEX maps of the vertical total electron content, TEC, for more than half a solar cycle, 1999–2008. The TEC values are extracted from the 600 grid points of the map at latitudes 60° N to 60° S with a step of 5° and longitudes 0° to 345° E with a step of 15° providing the data for 00:00 to 23:00 h of local time. The local effects of the solar radiant energy are filtered out by normalizing of the TEC in terms of the solar zenith angle χ at a particular time and the local noon value χ0. The degree of perturbation, DTEC, is computed as log of TEC relative to quiet reference median for 27 days prior to the day of observation. The W-index map is generated by segmentation of DTEC with the relevant thresholds specified earlier for foF2 so that 1 or −1 stands for the quiet state, 2 or −2 for the moderate disturbance, 3 or −3 for the moderate ionospheric storm, and 4 or −4 for intense ionospheric storm at each grid point of the map. The planetary ionospheric storm Wp index is obtained from the W-index map as a latitudinal average of the distance between maximum positive and minimum negative W-index weighted by the latitude/longitude extent of the extreme values on the map. The threshold Wp exceeding 4.0 index units and the peak value Wpmax≥6.0 specify the duration and the power of the planetary ionosphere-plasmasphere storm. It is shown that the occurrence of the Wp storms is growing with the phase of the solar cycle being twice as much as the number of the magnetospheric storms with Dst≤−100 nT and Ap≥100 nT.

3‐D eikonal solvers: First‐arrival traveltimes

Geophysics ◽  
2002 ◽  
Vol 67 (4) ◽  
pp. 1225-1231 ◽  
Author(s):  
Seongjai Kim
Keyword(s):  
Grid Point ◽  
Level Set Methods ◽  
Second Order ◽  
Normal Velocity ◽  
Fast Marching Method ◽  
Fast Marching ◽  
Maximum Angle ◽  
First Arrival ◽  
Grid Points ◽  
Marching Method

The article is concerned with the development and comparison of three different algorithms for the computation of first‐arrival traveltimes: the fast marching method (FMM), the group marching method (GMM), and a second‐order finite‐difference eikonal solver. GMM is introduced as a variant of FMM. It proceeds the solution by advancing a selected group of grid points at a time, rather than sorting the solution in the narrow band to march forward a single grid point. The second‐order eikonal solver studied in the article is an expanding‐box, essentially nonoscillatory scheme for which the stability is enforced by the introduction of a down ‘n’ out marching and a post‐sweeping iteration. Techniques such as the maximum angle condition, the average normal velocity, and cache‐based implementation are introduced for the algorithms to improve the numerical accuracy and efficiency. The algorithms are implemented for solving the eikonal equation in 3‐D isotropic media, and their performances are compared. GMM is numerically verified to be faster than FMM. However, the second‐order algorithm turns out to be superior to these first‐order level‐set methods in both accuracy and efficiency; the incorporation of average normal velocity improves accuracy dramatically for the second‐order scheme.

scholarly journals Estimating Salinity between 25° and 45°S in the Atlantic Ocean Using Local Regression

2008 ◽  
Vol 25 (1) ◽  
pp. 114-130 ◽  
Author(s):  
W. C. Thacker
Keyword(s):  
Grid Point ◽  
Empirical Relationship ◽  
Three Dimensional ◽  
Independent Data ◽  
Sufficient Data ◽  
Local Regression ◽  
Entire Region ◽  
Data Errors ◽  
Statistical Reliability ◽  
Grid Points

Abstract The empirical relationship between salinity and temperature in the South Atlantic is quantified with the aid of local regression. To capture the spatial character of the TS relationship, models are fitted to data for each point on a three-dimensional grid with spacing of 1° in latitude, 2° in longitude, and 25 dbar in the vertical. To ensure sufficient data for statistical reliability each fit is to data from a region extending over several grid points weighted so that more remote data exert less influence than those closer to the target grid point. Both temperature and its square are used as regressors to capture the curvature seen in TS plots, and latitude and longitude are used to capture systematic spatial variations over the fitting regions. In addition to using statistics of residuals to characterize how well the models fit the data, errors for data not used in fitting are examined to verify the models’ abilities to simulate independent data. The best model overall for the entire region at all depths is quadratic in temperature and linear in longitude and latitude.

scholarly journals Efficient, high-performance semantic segmentation using multi-scale feature extraction

PLoS ONE ◽  
2021 ◽  
Vol 16 (8) ◽  
pp. e0255397
Author(s):  
Moritz Knolle ◽  
Georgios Kaissis ◽  
Friederike Jungmann ◽  
Sebastian Ziegelmayer ◽  
Daniel Sasse ◽  
...  
Keyword(s):  
Deep Learning ◽  
Graphics Processing Units ◽  
Substantial Reduction ◽  
Image Features ◽  
Tumor Segmentation ◽  
Processing Unit ◽  
Central Processing ◽  
Multi Scale ◽  
Computational Performance ◽  
Wide Range

The success of deep learning in recent years has arguably been driven by the availability of large datasets for training powerful predictive algorithms. In medical applications however, the sensitive nature of the data limits the collection and exchange of large-scale datasets. Privacy-preserving and collaborative learning systems can enable the successful application of machine learning in medicine. However, collaborative protocols such as federated learning require the frequent transfer of parameter updates over a network. To enable the deployment of such protocols to a wide range of systems with varying computational performance, efficient deep learning architectures for resource-constrained environments are required. Here we present MoNet, a small, highly optimized neural-network-based segmentation algorithm leveraging efficient multi-scale image features. MoNet is a shallow, U-Net-like architecture based on repeated, dilated convolutions with decreasing dilation rates. We apply and test our architecture on the challenging clinical tasks of pancreatic segmentation in computed tomography (CT) images as well as brain tumor segmentation in magnetic resonance imaging (MRI) data. We assess our model’s segmentation performance and demonstrate that it provides performance on par with compared architectures while providing superior out-of-sample generalization performance, outperforming larger architectures on an independent validation set, while utilizing significantly fewer parameters. We furthermore confirm the suitability of our architecture for federated learning applications by demonstrating a substantial reduction in serialized model storage requirement as a surrogate for network data transfer. Finally, we evaluate MoNet’s inference latency on the central processing unit (CPU) to determine its utility in environments without access to graphics processing units. Our implementation is publicly available as free and open-source software.

Sign in / Sign up

Export Citation Format

Share Document