Convolutional Neural Network and Maxpooling Architecture on Zynq SoC FPGA

Bachtiar, Y.A.; Adiono, Trio

doi:10.1109/isesd.2019.8909510

Cited by 14 publications

(6 citation statements)

References 4 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The previous implementation also proves that the representation ability of networks can be promoted significantly by appealing both average-pooled and max-pooled features simultaneously [40]. Average pooling and max pooling are sample-based discretization process, in which average pooling [41] uses average value of the filter while max pooling [42] uses max value. Therefore, we employ both technologies in channel attention:…”

Section: Attention Modulementioning

confidence: 77%

SuperMeshing: A New Deep Learning Architecture for Increasing the Mesh Density of Physical Fields in Metal Forming Numerical Simulation

Nie

et al. 2021

Journal of Applied Mechanics

View full text Add to dashboard Cite

In stress field analysis, the finite element method is a crucial approach, in which the mesh-density has a significant impact on the results. High mesh density usually contributes authentic to simulation results but costs more computing resources. To eliminate this drawback, we propose a data-driven mesh-density boost model named SuperMeshingNet that uses low mesh-density as inputs, to acquire high-density stress field instantaneously, shortening computing time and cost automatically. Moreover, the Res-UNet architecture and attention mechanism are utilized, enhancing the performance of SuperMeshingNet. Compared with the baseline that applied the linear interpolation method, SuperMeshingNet achieves a prominent reduction in the mean squared error (MSE) and mean absolute error (MAE) on the test data. The well-trained model can successfully show more excellent performance than the baseline models on the multiple scaled mesh-density, including 2X, 4X, and 8X. Enhanced by SuperMeshingNet with broaden scaling of mesh density and high precision output, FEA can be accelerated with seldom computational time and cost.

show abstract

Section: Attention Modulementioning

confidence: 77%

SuperMeshing: A New Deep Learning Architecture for Increasing the Mesh Density of Physical Fields in Metal Forming Numerical Simulation

Nie

et al. 2021

Journal of Applied Mechanics

View full text Add to dashboard Cite

show abstract

“…Y. A. Bachtiar and T. Adiono proposed a configurable CNN and maximum pool processor architecture based on a low-capacity SoC, and completed the algorithm validation on Xilinx Zynq-7000 SoC [25]. In order to reduce the calculation time, S. Ghaffari and S. Sharifian proposed a new method to calculate the hyperbolic tangent activation function, and implemented the LeNet convolutional network architecture and used MNIST data sets as samples to identify handwritten numerals [26].…”

Section: Related Workmentioning

confidence: 99%

Research on the Lightweight Deployment Method of Integration of Training and Inference in Artificial Intelligence

Zheng

2022

Applied Sciences

View full text Add to dashboard Cite

In recent years, the continuous development of artificial intelligence has largely been driven by algorithms and computing power. This paper mainly discusses the training and inference methods of artificial intelligence from the perspective of computing power. To address the issue of computing power, it is necessary to consider performance, cost, power consumption, flexibility, and robustness comprehensively. At present, the training of artificial intelligence models mostly are based on GPU platforms. Although GPUs offer high computing performance, their power consumption and cost are relatively high. It is not suitable to use GPUs as the implementation platform in certain application scenarios with demanding power consumption and cost. The emergence of high-performance heterogeneous architecture devices provides a new path for the integration of artificial intelligence training and inference. Typically, in Xilinx and Intel’s multi-core heterogeneous architecture, multiple high-performance processors and FPGAs are integrated into a single chip. When compared with the current separate training and inference method, heterogeneous architectures leverage a single chip to realize the integration of AI training and inference, providing a good balance of training and inference of different targets, further reducing the cost of training and implementation of AI inference and power consumption, so as to achieve the lightweight goals of computation, and to improve the flexibility and robustness of the system. In this paper, based on the LeNet-5 network structure, we first introduced the process of network training using a multi-core CPU in Xilinx’s latest multi-core heterogeneous architecture device, MPSoC. Then, the method of converting the network model into hardware logic implementation was studied, and the model parameters were transferred from the processing system of the device to the hardware accelerator structure, composed of programmable logic through the bus interface AXI provided on the chip. Finally, the integrated implementation method was tested and verified in Xilinx MPSoC. According to the test results, the recognition accuracy of this lightweight deployment scheme on MNIST dataset and CIFAR-10 dataset reached 99.5 and 75.4% respectively, while the average processing time of the single frame was only 2.2 ms. In addition, the power consumption of the network within the SoC hardware accelerator is only 1.363 W at 100 MHz.

show abstract

“…The initialized number of separable convolutional filters are 32 and by the factor of 2 they are increased in successive convolutional layers. These separable convolutional layers are followed by MaxPooling2D and it is used to reduce the dimensions of the input data and allows us to make assumptions for features binned in the sub-regions, and 2D means the dimensions of the input data is in two dimensions like (x-axis, y-axis) [25]. And to optimize the parameters we have used Adam optimizer [26].…”

Section: A Covid-19 Detection Modelmentioning

confidence: 99%

Research on Data Science Ensembles for Covid-19 Detection and Length of Stay Prediction

Sinha

Tushar

Goel

2021

2021 International Conference on Computing, Communication, and Intelligent Systems (ICCCIS)

View full text Add to dashboard Cite

We have demonstrated the use of an iteratively severed model of deep learning which associates for diagnosing Covid-19 pulmonary demonstration of using chest X-rays. In this paper, a customized convolutional neural network model is trained and analyzed on publicly available chest X-rays to grasp modality-strict feature demonstrations. Since the best performing models learn iteratively to make the model memory efficient, this model also learns and tries to improve the results with each step and classify the chest X-rays in their categories accurately. Then another model which predicts the length of stay of a patient at the hospital is created using multi-layered data processing approach. This model will empower hospitals for on time interference to prevent confusions and better management of hospital resources. We propose a method that uses catboost model which generally classifies the data in multiple classes. As a result, this study provides modality strict iterative and knowledge reusable model which influences Covid-19 detection and length of stay prediction.

show abstract

Convolutional Neural Network and Maxpooling Architecture on Zynq SoC FPGA

Cited by 14 publications

References 4 publications

SuperMeshing: A New Deep Learning Architecture for Increasing the Mesh Density of Physical Fields in Metal Forming Numerical Simulation

SuperMeshing: A New Deep Learning Architecture for Increasing the Mesh Density of Physical Fields in Metal Forming Numerical Simulation

Research on the Lightweight Deployment Method of Integration of Training and Inference in Artificial Intelligence

Research on Data Science Ensembles for Covid-19 Detection and Length of Stay Prediction

Contact Info

Product

Resources

About