6,276 research outputs found
Elastic-DF: Scaling Performance of DNN Inference in FPGA Clouds through Automatic Partitioning
No full textCustomized compute acceleration in the datacenter is key to the wider roll-out of applications based on deep neural network (DNN) inference. In this article, we investigate how to maximize the performance and scalability of field-programmable gate array (FPGA)-based pipeline dataflow DNN inference accelerators (DFAs) automatically on computing infrastructures consisting of multi-die, network-connected FPGAs. We present Elastic-DF, a novel resource partitioning tool and associated FPGA runtime infrastructure that integrates with the DNN compiler FINN. Elastic-DF allocates FPGA resources to DNN layers and layers to individual FPGA dies to maximize the total performance of the multi-FPGA system. In the resulting Elastic-DF mapping, the accelerator may be instantiated multiple times, and each instance may be segmented across multiple FPGAs transparently, whereby the segments communicate peer-to-peer through 100 Gbps Ethernet FPGA infrastructure, without host involvement. When applied to ResNet-50, Elastic-DF provides a 44% latency decrease on Alveo U280. For MobileNetV1 on Alveo U200 and U280, Elastic-DF enables a 78% throughput increase, eliminating the performance difference between these cards and the larger Alveo U250. Elastic-DF also increases operating frequency in all our experiments, on average by over 20%. Elastic-DF therefore increases performance portability between different sizes of FPGA and increases the critical throughput per cost metric of datacenter inference. Green Open Access added to TU Delft Institutional Repository 'You share, we take care!' - Taverne project https://www.openaccess.nl/en/you-share-we-take-care Otherwise as indicated in the copyright section: the publisher is the copyright holder of this work and the author uses the Dutch legislation to make this work public.Computer Engineerin
First-order gradiometer of high Tc rf SQUID
No full textWe developed a first-order gradiometer by subtracting the outputs of two sets of rf SQUIDs with 6 cm base distance between two SQUIDs. When the gradiometer was balanced by only tuning a potentiometer, the common mode rejection ratio (CMRR) was about 640. The flux noise spectrum of gradiometer in laboratory without shielding was in the same order of magnitude as that of a magnetometer in shielding. The gradient resolution of the gradiometer was 53fT/(root Hz.cm). We used a 0.08-45 Hz bandpass filter and 50 Hz, 100 Hz, 150 Hz notch filters to suppress the environmental noise and interferences of 50 Hz line frequency and its harmonics. The gradiometer could detect the magnetocardiogram (MCG) in laboratory.Physics, AppliedSCI(E)EI
Data Ellipses, HE Plots and Reduced-Rank Displays for Multivariate Linear Models: SAS Software and Examples
No full textThis paper describes graphical methods for multiple-response data within the framework of the multivariate linear model (MLM), aimed at understanding what is being tested in a multivariate test, and how factor/predictor effects are expressed across multiple response measures. In particular, we describe and illustrate a collection of SAS macro programs for: (a) Data ellipses and low-rank biplots for multivariate data, (b) HE plots, showing the hypothesis and error covariance matrices for a given pair of responses, and a given effect, (c) HE plot matrices, showing all pairwise HE plots, and (d) low-rank analogs of HE plots, showing all observations, group means, and their relations to the response variables.
Hydrodynamical turbulence by fractal fourier decimation
No full textWe present a systematic numerical investigation of high-resolution 3D isotropic and homogeneous turbulence resolved on a decimated set of Fourier modes. Fractal decimation acts to decrease the effective dimensionality of the flow by allowing triadic interactions only in a set of Fourier modes N(k) proportional to k^DF for large k. While keeping the symmetries of the original 3D Navier-Stokes equations unchanged, a dramatic change in small-scale statistics is detected at decreasing the fractal dimension DF . Already at fractal dimension DF = 2.8, a global self-similar behaviour is observed in the inertial range of scales, the consequence of such transition are the restoration of the scaling symmetry and vorticity distribution that becomes close to Gaussian. We relate the results to the different roles of local vs non-local interactions in the energy transfer range
DF-2 septicemia following whirlpool spa immersion.
No full textWe describe the case of a 31-year-old asplenic man who developed DF-2 bacteremia, septic shock, and pneumonia after recreational immersion in a whirlpool spa. The patient did not have a history of dog bite or contact with canine secretions, although he owned two dogs. DF-2 could not be isolated from the whirlpool spa
'Response by the author, Daniel F. Vukovich.'
No full textResponse by the author (Vukovich) to a review of Illiberal China (my 2019 monograph
DF AND HF LASER SPECTRA
No full textW. B. Roh and K. Narahari Rao, J. Mol. Spectrosc. 49, 317 (1974).""Author Institution: Department of Physics, The Ohio State University; Department of Electrical Engineering, The Ohio State UniversityA pulsed laser was used as a source to study DF and HF infrared laser spectra with a grating spectrometer. The rotational structure observed for the 1-0, 2-1, and 3-2 bands of DF, and the 1-0 and 2-1 bands of HF has been measured by following a procedure similar to that used for the CO laser Molecular constants derived by combining these laser measurements with other available data will be presented
Retinitis Pigmentosa GTPase Regulator (RPGR) protein isoforms in mammalian retina:insights into X-linked Retinitis Pigmentosa and associated ciliopathies
No full textMutations in the cilia-centrosomal protein Retinitis Pigmentosa GTPase Regulator (RPGR) are a frequent cause of retinal degeneration. The RPGR gene undergoes complex alternative splicing and encodes multiple protein isoforms. To elucidate the function of major RPGR isoforms (RPGR 1-19 and RPGR ORF15), we have generated isoform-specific antibodies and examined their expression and localization in the retina. Using sucrose-gradient centrifugation, immunofluorescence and co-immunoprecipitation methods, we show that RPGR isoforms localize to distinct sub-cellular compartments in mammalian photoreceptors and associate with a number of cilia-centrosomal proteins. The RCC1-like domain of RPGR, which is present in all major RPGR isoforms, is sufficient to target it to the cilia and centrosomes in cultured cells. Our findings indicate that multiple isotypes of RPGR may perform overlapping yet somewhat distinct transport-related functions in photoreceptors
OPTOACOUSTIC MEASUREMENT OF DF LASER ABSORPTION BY METHANE
No full textAuthor Institution:An optoacoustic system has been constructed for measurement of absorption of DF laser radiation. Absorption by methane in Argon and Nitrogen has been measured at 15 DF laser lines in the 3.6 – 4.0 region. Construction and calibration of the spectrophone will be discussed
A Synergistic CNN-DF Method for Landslide Susceptibility Assessment
No full textThe complex structures and intricate hyperparameters of existing deep learning (DL) models make achieving higher accuracy in landslide susceptibility assessment (LSA) time-consuming and labor-intensive. Deep forest (DF) is a decision tree-based DL framework that uses a cascade structure to process features, with model depth adapting to the input data. To explore a more ideal landslide susceptibility model, this study designed a landslide susceptibility model combining convolutional neural networks (CNNs) and DF, referred to as CNN-DF. The Bailong River Basin, a region severely affected by landslides, was chosen as the study area. First, the landslide inventory and influencing factors of the study area were obtained. Second, an equal number of landslide and nonlandslide samples were selected under similar environmental constraints to establish the dataset. Third, CNN was used to extract high-level features from the raw data, which were then input into the DF model for training and testing. Finally, the trained model was used to predict landslide susceptibility. The results showed that the CNN-DF model achieved high prediction accuracy, with an AUC of 0.9061 on the testing set, outperforming DF, CNN, and other commonly used machine learning models. In landslide susceptibility maps (LSMs), the proportion of historical landslides in the very high susceptibility category of CNN-DF was also higher than that of other models. CNN-DF is feasible for LSA, offering higher efficiency and more accurate results. In addition, the SHAP algorithm was used to quantify the contribution of features to the prediction results both globally and locally, further explaining the model. The LSM based on CNN-DF can provide a scientific basis for landslide prevention and disaster management in the target area
- …
