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.

Study of the dynamic response of a CAARC building by the synthetic wind method and comparison with different approaches

Purpose This paper analyzes the effect that is generated in the dynamic response of a Commonwealth Advisory Aeronautical Council building for different types of power spectra. This article also compares synthetic wind method (SWM) results with wind tunnel tests and other numerical approaches.Design/methodology/approach One of the main methodologies developed in Brazil, the SWM, is employed to determine the dynamic wind loads. The Davenport, Lumley and Panowski, Harris, von Karman and Kaimal model are used in SWM to generate the resonant harmonics. Lateral pressures are calculated by the wind speed deflection profile for 30, 35, 40 and 45 m/s. The structure is processed in Autodesk Robot Structural Analysis with numerical analysis in FEM by the Hilber–Hughes–Taylor method. To corroborate the synthetic wind with experimental results, displacement curves are developed for wind tunnel experimental results, Davenport method, Eurocode and NBR 6123, together with the SWM.Findings Results show that for 30 m/s, the lowest convergence of the power spectra models was presented and that the greatest difference found was below 10%. In addition, it was shown that Eurocode 1-4 can lead to oversizing, while NBR 6123 can lead to undersizing, compared with the experimental results. Finally, results by the Davenport method, wind tunnel test and synthetic wind showed good accuracy.Originality/value By carrying out this comparative analysis, this work presents an important contribution in the field of calculating the dynamic response of tall buildings. Studies with these comparisons to corroborate the SWM had not yet been carried out.

CacheHawkeye: Detecting Cache Side Channel Attacks Based on Memory Events

Cache side channel attacks, as a type of cryptanalysis, seriously threaten the security of the cryptosystem. These attacks continuously monitor the memory addresses associated with the victim’s secret information, which cause frequent memory access on these addresses. This paper proposes CacheHawkeye, which uses the frequent memory access characteristic of the attacker to detect attacks. CacheHawkeye monitors memory events by CPU hardware performance counters. We proved the effectiveness of CacheHawkeye on Flush+Reload and Flush+Flush attacks. In addition, we evaluated the accuracy of CacheHawkeye under different system loads. Experiments demonstrate that CacheHawkeye not only has good accuracy but can also adapt to various system loads.

Fake News Detection Using NLP

The spreading of fake news has given rise to many problems in society. It is due to its ability to cause a lot of social and national damage with destructive impacts. Sometimes it gets very difficult to know if the news is genuine or fake. Therefore it is very important to detect if the news is fake or not. "Fake News" is a term used to represent fabricated news or propaganda comprising misinformation communicated through traditional media channels like print, and television as well as nontraditional media channels like social media. Techniques of NLP and Machine learning can be used to create models which can help to detect fake news. In this paper we have presented six LSTM models using the techniques of NLP and ML. The datasets in comma-separated values format, pertaining to political domain were used in the project. The different attributes like the title and text of the news headline/article were used to perform the fake news detection. The results showed that the proposed solution performs well in terms of providing an output with good accuracy, precision and recall. The performance analysis made between all the models showed that the models which have used GloVe and Word2vec method work better than the models using TF-IDF. Further, a larger dataset for better output and also other factors such as the author ,publisher of the news can be used to determine the credibility of the news. Also, further research can also be done on images, videos, images containing text which can help in improving the models in future. Keywords: Fake news detection, LSTM(long short term memory),Word2Vec,TF-IDF,Natural Language Processing.

Empirical Correlation of Emulsion Size Prediction for Zinc Extraction Using Flat Blade Impeller System in Emulsion Liquid Membrane Process

In this study, determination of droplets in the presence of blended mixture of surfactants (Span 80 and Tween 80) and nanoparticles, iron (III) oxide (Fe2O3) were investigated using a single stage mixer-settler extractor with 4-pitched flat blade impeller on one shaft employment. Additionally, the influence of Fe2O3 and blended surfactant mixture of Span 80 and Tween 80 on the dispersion of emulsion in terms of Sauter diameter (D32) measurement was compared with new correlations. Results indicate that the presence of Fe2O3 in the blended mixture of surfactant simultaneously decreased in D32 by 79 % and the stability of the emulsion system was enhanced. Overall, empirical correlation for droplet size at different conditions are obtained, and the modified correlation for D32 is presented. The correlation found is D32/DI =0.02265(3.419Φi−1)We-0.6. The calculated average absolute relative deviation (%AARD) is 2.69 %, thus indicating a good accuracy and acceptability between the presented correlation and experimental data.

Multi-jet ice 3D printing

Purpose Multi-jet deposition of the materials is a matured technology used for graphic printing and 3 D printing for a wide range of materials. The multi-jet technology is fine-tuned for liquids with a specific range of viscosity and surface tension. However, the use of multi-jet for low viscosity fluids like water is not very popular. This paper aims to demonstrate the technique, particularly for the water-ice 3 D printing. 3 D printed ice parts can be used as patterns for investment casting, templates for microfluidic channel fabrication, support material for polymer 3 D printing, etc. Design/methodology/approach Multi-jet ice 3 D printing is a novel technique for producing ice parts by selective deposition and freezing water layers. The paper confers the design, embodiment and integration of various subsystems of multi-jet ice 3 D printer. The outcomes of the machine trials are reported as case studies with elaborate details. Findings The prismatic geometries are realized by ice 3 D printing. The accuracy of 0.1 mm is found in the build direction. The part height tends to increase due to volumetric expansion during the phase change. Originality/value The present paper gives a novel architecture of the ice 3 D printer that produces the ice parts with good accuracy. The potential applications of the process are deliberated in this paper.

Using Ground Reaction Forces To Predict Fatigue In Treadmill Running: A Machine Learning Approach

The analysis of running patterns, especially those associated with fatigue, can help specialists in designing more efficient workouts and preventing injuries in high-performance sports. However, classifying running patterns is not trivial for humans. An interesting alternative is to use Machine Learning methods, such as Artificial Neural Networks (ANNs), to classify running patterns. In this work, ground reaction forces are measured by sensors coupled to the base of a low-cost open-source treadmill. ANNs are used to classify the force signals and to indicate the occurrence of fatigue. Different features, extracted from the force signals, are proposed and investigated. A Genetic Algorithm (GA) is used to select the best features. The experimental results indicate that the ANN is able to classify the running patterns with good accuracy. In addition, some features selected by the GA provide important information regarding the identification of fatigue in treadmill running.

A Simple and Sensitive Inhibitory Kinetic Method for the Carbocisteine Determination

A fast, reproducible, and sensitive method is proposed for the kinetic determination of carbocisteine (CCys). The method depends on the inhibitory property of carbocisteine, which reduces the Hg2+ catalyzed substitution rate of cyanide from [Ru(CN)6]4- with N-R-salt (1-Nitroso-2-naphthol-3,6-disulfonic acid disodium salt) via forming a stable complex with Hg2+. Spectrophotometric measurements were carried out by recording the absorbance at 525 nm (λmax of [Ru(CN)5 Nitroso-R-Salt]3- complex) at a fixed time of 10 and 15 min under the optimized reaction conditions with [N-R-salt] = 4.5 × 10-4 M, I = 0.05 M (KNO3), Temp = 45.0 ± 0.2 o C, pH = 7.0 ± 0.03, [Hg2+] = 8.0 × 10-5 M and [Ru(CN)64-] = 4.25 × 10-5 M. With the proposed method, CCys can be determined quantitatively down to 3.0 × 10-6 M. This methodology can be effectively used for the rapid quantitative estimation of CCys in the pharmaceutical samples with good accuracy and reproducibility. The addition of common excipients in pharmaceuticals even up to 1000 times with [CCys] does not interfere significantly in the estimation of CCys. Resumen. Se propone un método rápido, reproducibley sensible para la determinación cinética de la carbocisteina (CCys). El método depende de la propiedad inhibitoria de la carbocisteina que reduce la tasa de sustitución catalizada por Hg2+ del cianuro de [Ru(CN)6]4- con la sal N-R (sal disódica del ácido 1-Nitroso-2-naftol-3,6-disulfónico) mediante la formación de un complejo estable con Hg2+. Las mediciones espectrofotométricas se llevaron a cabo registrando la absorbancia a 525 nm (λmax del complejo [Ru(CN)5 Sal-Nitroso-R]3-) en un tiempo fijo de 10 y 15 min en las condiciones de reacción optimizadas con [sal-NR] = 4.5 × 10-4 M, I = 0.05 M (KNO3), Temp = 45.0 ± 0.2 o C, pH = 7.0 ± 0.03, [Hg2+] = 8.0 × 10-5 M y [Ru(CN)64-] = 4.25 × 10-5 M. Con el método propuesto, CCys se puede determinar cuantitativamente hasta 3,0 × 10-6 M. Esta metodología se puede utilizar eficazmente para la estimación cuantitativa rápida de CCys en las muestras farmacéuticas con buena precisión y reproducibilidad. La adición de excipientes comunes en productos farmacéuticos incluso hasta 1000 veces con [CCys] no interfiere significativamente en la estimación de CCys.

Modeling Asymmetry of a Current–Voltage Curve of a Novel MF-4SC/PTMSP Bilayer Membrane

A novel bilayer cation-exchange membrane—consisting of a thick layer of a pristine perfluorinated membrane MF-4SC (Russian equivalent of Nafion®-117) and a thinner layer (1 μm) of the membrane, on a base of glassy polymer of internal microporosity poly(1-trimethylsilyl-1-propyne) (PTMSP)—was prepared and characterized. Using the physicochemical characteristics of one-layer membranes MF-4SC and PTMSP in 0.05 M HCl and NaCl solutions, the asymmetric current–voltage curves (CVC) of the bilayer composite were described with good accuracy up to the overlimiting regime, based on the “fine-porous membrane” model. The MF-4SC/PTMSP bilayer composite has a significant asymmetry of CVC that is promising for using it in electromembrane devices, such as membrane detectors, sensors, and diodes.

Classification of Plant Leaves Using New Compact Convolutional Neural Network Models

Precision crop safety relies on automated systems for detecting and classifying plants. This work proposes the detection and classification of nine species of plants of the PlantVillage dataset using the proposed developed compact convolutional neural networks and AlexNet with transfer learning. The models are trained using plant leaf data with different data augmentations. The data augmentation shows a significant improvement in classification accuracy. The proposed models are also used for the classification of 32 classes of the Flavia dataset. The proposed developed N1 model has a classification accuracy of 99.45%, N2 model has a classification accuracy of 99.65%, N3 model has a classification accuracy of 99.55%, and AlexNet has a classification accuracy of 99.73% for the PlantVillage dataset. In comparison to AlexNet, the proposed models are compact and need less training time. The proposed N1 model takes 34.58%, the proposed N2 model takes 18.25%, and the N3 model takes 20.23% less training time than AlexNet. The N1 model and N3 models are size 14.8 MB making it 92.67% compact, and the N2 model is 29.7 MB which makes it 85.29% compact as compared to AlexNet. The proposed models are giving good accuracy in classifying plant leaf, as well as diseases in tomato plant leaves.