Low-precision Floating-point Arithmetic for High-performance FPGA-based CNN Acceleration

Low-precision data representation is important to reduce storage size and memory access for convolutional neural networks (CNNs). Yet, existing methods have two major limitations: (1) requiring re-training to maintain accuracy for deep CNNs and (2) needing 16-bit floating-point or 8-bit fixed-point for a good accuracy. In this article, we propose a low-precision (8-bit) floating-point (LPFP) quantization method for FPGA-based acceleration to overcome the above limitations. Without any re-training, LPFP finds an optimal 8-bit data representation with negligible top-1/top-5 accuracy loss (within 0.5%/0.3% in our experiments, respectively, and significantly better than existing methods for deep CNNs). Furthermore, we implement one 8-bit LPFP multiplication by one 4-bit multiply-adder and one 3-bit adder, and therefore implement four 8-bit LPFP multiplications using one DSP48E1 of Xilinx Kintex-7 family or DSP48E2 of Xilinx Ultrascale/Ultrascale+ family, whereas one DSP can implement only two 8-bit fixed-point multiplications. Experiments on six typical CNNs for inference show that on average, we improve throughput by over existing FPGA accelerators. Particularly for VGG16 and YOLO, compared to six recent FPGA accelerators, we improve average throughput by 3.5 and 27.5 and average throughput per DSP by 4.1 and 5 , respectively.

A Roadway Safety Sustainable Approach: Modeling for Real-Time Traffic Crash with Limited Data and Its Reliability Verification

With the development of freeway system informatization, it is easier to obtain the traffic flow data of freeway, which are widely used to study the relationship between traffic flow state and traffic safety. However, as the development degree of the freeway system is different in different regions, the sample size of traffic data collected in some regions is insufficient, and the precision of data is relatively low. In order to study the influence of limited data on the real-time freeway traffic crash risk modeling, three data sets including high precision data, small sample data, and low precision data were considered. Firstly, Bayesian Logistic regression was used to identify and predict the risk of three data sets. Secondly, based on the Bayesian updating method, the migration test towards high and low precision data sets was established. Finally, the applicability of machine learning and statistical methods to low precision data set was compared. The results show that the prediction performance of Bayesian Logistic regression improves with the increasing of sample size. Bayesian Logistic regression can identify various significant risk factors when data sets are of different precision. Comparatively, the prediction performance of the support vector machine is better than that of Bayesian Logistic. In addition, Bayesian updating method can improve the prediction performance of the transplanted model.

Experimental Review of ΛΛ¯ Production

Exclusive hyperon-antihyperon production provides a unique insight for understanding of the intrinsic dynamics when strangeness is involved. In this paper, we review the results of ΛΛ¯ production via different reactions from various experiments, e.g., via p¯p annihilation from the LEAR experiment PS185, via electron-positron annihilation using the energy scan method at the CLEO-c and BESIII experiments and the initial-state-radiation approach utilized at the BaBar experiment. The production cross section of ΛΛ¯ near the threshold is sensitive to QCD based prediction. Experimental high precision data for p¯p→Λ¯Λ close to the threshold region is obtained. The cross section of e+e−→ΛΛ¯ is measured from its production threshold to high energy. A non-zero cross section for e+e−→ΛΛ¯ near threshold is observed at BaBar and BESIII, which is in disagreement with the pQCD prediction. However, more precise data is needed to confirm this observation. Future experiments, utilizing p¯p reaction such as PANDA experiment or electron-positron annihilation such as the BESIII and BelleII experiments, are needed to extend the experimental data and to understand the ΛΛ¯ production.

Content validation to support the monitoring of disclosure of HIV diagnosis in childhood

ABSTRACT Objectives: to create and validate the content of a guide for monitoring the communication of the HIV diagnosis in childhood. Methods: methodological study, with a design guided by the Knowledge-to-Action (KTA) Framework, supported by a participatory approach. The guide’s content was structured according to the communication elements proposed by Lasswel from review studies. Results: the content was validated by 26 experts from nursing, medicine, psychology and pedagogy, using a Likert-type scale for relevance, clarity and precision. Data collection took place online and achieved a Content Validity Index of 0.94. Conclusions: the guide can contribute to the practice of professionals who care for children living with HIV, to support the family in communication and to the child’s right to know their diagnosis. For further research, it is recommended to create and validate the face of the technology in order to implement it.

Delineation of Management Zones in Hedgerow Almond Orchards Based on Vegetation Indices from UAV Images Validated by LiDAR-Derived Canopy Parameters

One of the challenges in orchard management, in particular of hedgerow tree plantations, is the delineation of management zones on the bases of high-precision data. Along this line, the present study analyses the applicability of vegetation indices derived from UAV images to estimate the key structural and geometric canopy parameters of an almond orchard. In addition, the classes created on the basis of the vegetation indices were assessed to delineate potential management zones. The structural and geometric orchard parameters (width, height, cross-sectional area and porosity) were characterized by means of a LiDAR sensor, and the vegetation indices were derived from a UAV-acquired multispectral image. Both datasets summarized every 0.5 m along the almond tree rows and were used to interpolate continuous representations of the variables by means of geostatistical analysis. Linear and canonical correlation analyses were carried out to select the best performing vegetation index to estimate the structural and geometric orchard parameters in each cross-section of the tree rows. The results showed that NDVI averaged in each cross-section and normalized by its projected area achieved the highest correlations and served to define potential management zones. These findings expand the possibilities of using multispectral images in orchard management, particularly in hedgerow plantations.

A New Second-Derivative Spectrophotometric Method for the Determination of Vildagliptin in Pharmaceutical dosage form

An accurate derivative spectrophotometric method was developed and validated for the determination of dipeptidyl peptidase inhibitor vildagliptin in the pharmaceutical dosage form. The second derivative of the UV spectra has enabled the estimation of vildagliptin absorbance at 217 nm without any interferences. Linearity, precision, accuracy, detection (LOD), and quantification (LOQ) limits were established for method validation. Calibration curve was linear in the range of 10-60 µg/mL with a regression coefficient of 0.998. The method was validated as per the International Conference on Harmonization (ICH Q2 (R1)). The limit of detection and the limit of quantification were found to be 2.06 µg/mL and 6.25 µg/mL, respectively. Intra and interday precision data illustrated that the method has acceptable reproducibility as the percentage relative standard deviation (RSD) was less than 2 %, which indicates the precision of the method. The recovery was 98.39 % by the standard addition method. The percentage assay of vildagliptin was 98.06 % which showed good applicability. The following results indicate that the procedure is accurate, precise, and reproducible while being simple and less time-consuming. The method was demonstrated to be adequate for routine analysis in quality control.

A Corpus-Based and Complex Computing Digital Media System for 3D Animation

In this paper, we design a corpus-based 3D animation digital media system to improve the accuracy of 3D animation generation and realize crossplatform animation display. The corpus module extracts high-precision data through web crawling, web cleaning, Chinese word separation, and text classification steps; the character animation generation module uses the semantic description method to expand the frame information description of the extracted data, calculates the object spatial 3D coordinates, and uses the built-in animation execution script to generate 3D character animation; the improved digital media player module uses the improved digital media player to realize crossplatform display of 3D character animations using the improved digital media player. By constructing multidimensional character relationships and combining multiple visualization methods, the complex and multifaceted social relationship network is made available to users in an intuitive and more acceptable and understandable mode. Through a large number of user surveys, it is proved that the visual analysis method combining real social and virtual social proposed in this paper provides a more adequate and reliable basis for friend recommendation and social network analysis; the combination of multiple character relationships with geographical information and the use of visualization to describe multidimensional historical character relationships provides a new research perspective for the research and exploration of humanistic neighborhoods. The experimental results prove that the designed system can effectively read known contents and extract keywords and generate 3D animation based on keyword features, with a high accuracy rate, fast response time, small frame loss rate, and crossplatform display animation advantages.

FPGA-Based Convolutional Neural Network Accelerator with Resource-Optimized Approximate Multiply-Accumulate Unit

Convolutional neural networks (CNNs) are widely used in modern applications for their versatility and high classification accuracy. Field-programmable gate arrays (FPGAs) are considered to be suitable platforms for CNNs based on their high performance, rapid development, and reconfigurability. Although many studies have proposed methods for implementing high-performance CNN accelerators on FPGAs using optimized data types and algorithm transformations, accelerators can be optimized further by investigating more efficient uses of FPGA resources. In this paper, we propose an FPGA-based CNN accelerator using multiple approximate accumulation units based on a fixed-point data type. We implemented the LeNet-5 CNN architecture, which performs classification of handwritten digits using the MNIST handwritten digit dataset. The proposed accelerator was implemented, using a high-level synthesis tool on a Xilinx FPGA. The proposed accelerator applies an optimized fixed-point data type and loop parallelization to improve performance. Approximate operation units are implemented using FPGA logic resources instead of high-precision digital signal processing (DSP) blocks, which are inefficient for low-precision data. Our accelerator model achieves 66% less memory usage and approximately 50% reduced network latency, compared to a floating point design and its resource utilization is optimized to use 78% fewer DSP blocks, compared to general fixed-point designs.

Water vapor estimation based on one-year data of E-band millimeter-wave link in the northeast of China

The amount of water vapor in the atmosphere is very small, but its content varies greatly in different humidity areas. The change of water vapor will affect the transmission of microwave link signals, and most of the water vapor is concentrated in the lower layer, so the water vapor density can be measured by the change of the near-ground microwave link transmission signal. This study collected one-year data of the E-band millimeter-wave link in Hebei, China, and used a model based on the ITU-R to estimate the water vapor density. An improved method of extracting the water vapor induced attenuation value is also introduced. It has a higher time resolution and the estimation error is lower than the previous method. In addition, this paper conducts the seasonal analysis of water vapor inversion for the first time. The monthly and seasonal evaluation index results show a high correlation between the retrieved water vapor density the actual water vapor density value measured by the local weather station. The correlation value for the whole year is up to 0.95, the root mean square error is as low as 0.35, and the average relative error is as low as 0.05. This research shows that millimeter-wave backhaul link provides high-precision data for the measurement of water vapor density and has a positive effect on future weather forecast research.

Validated Modernized Assay for Foscarnet in Pharmaceutical Formulations Using Suppressed Ion Chromatography Developed through a Quality by Design Approach

Inspired by the United States Pharmacopoeia (USP) “monograph modernization” initiative, we developed and validated an assay for foscarnet sodium injection solution (“foscavir”), following quality by design (QbD) principles, incorporating design of experiments (DoE) and multivariate data analysis to establish the design space and robust setpoint of the method. The resulting analytical procedure was based on ion chromatography (IC) with suppressed conductivity detection, employing an isocratic carbonate–bicarbonate eluent system. The assay was successfully validated at the robust setpoint conditions, according to the guidelines established by the International Council for Harmonization (ICH). The linear range stretched at least from 5 to 100 mg/L with high repeatability (relative standard deviation, RSD ≤ 0.3%) both at the target concentration (60 mg/L) and at 50% and 150% from this level. Special attention was given to establish a rugged assay that would be easily transferable between laboratories, and the recorded recoveries of 98.2–100.5% for both the formulated drug product and the drug substance during intermediate precision evaluation at different analysis situations indicated that this mission was accomplished. A multivariate assessment of intermediate precision data acquired using an experimental design scheme revealed that the assay was not adversely affected by any of the situation variables, including the use of different liquid chromatography instrument types, regardless of if they were constructed from inert materials or stainless steel that had been passivated, even though such problems have been reported in several previous methods for analysis of foscarnet.