Complex convolution Kernel for deep networks

Li, Kaizhou; Shi, Hong; Hu, Qinghua

doi:10.1109/wcsp.2016.7752631

Cited by 4 publications

(5 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The input of the convolution layer is data of A × B × C, where A × B is the data dimension of label in experiment, and C is the number of channels; that is, the dimension of the rag's data characteristics, and the data of the tags have three channels, which are the characteristics RSSI, Phase, and Timestamp, as shown in Figure 4. The convolutional layer has k convolution kernels of size m × m. These convolution kernels determine the area size of each neuron in the convolutional layer and the adjacent neurons in the previous layer [28]. When the neural network works, it regularly scans the characteristics of the input data.…”

Section: Mining Datamentioning

confidence: 99%

“…The input data is multiplied by the corresponding value in the area where the convolution kernel coincides, and then the sum is added to bias, and finally we get a value of the output data. The convolutional layer has k convolution kernels of size m × m. These convolution kernels determine the area size of each neuron in the convolutional layer and the adjacent neurons in the previous layer [28]. When the neural network works, it regularly scans the characteristics of the input data.…”

Section: Mining Datamentioning

confidence: 99%

“…Through the convolutional neural network, we obtained the information confidence of the z-axis direction of the labelled tag; the maximum value of the confidence of the position information of the tag is taken as the position of the tag in the Z-axis direction. The convolutional layer has k convolution kernels of size m × m. These convolution kernels determine the area size of each neuron in the convolutional layer and the adjacent neurons in the previous layer [28]. When the neural network works, it regularly scans the characteristics of the input data.…”

Section: Mining Datamentioning

confidence: 99%

See 2 more Smart Citations

3DLRA: An RFID 3D Indoor Localization Method Based on Deep Learning

Cheng

Wang

Guan

et al. 2020

Sensors

View full text Add to dashboard Cite

As the core supporting technology of the Internet of Things, Radio Frequency Identification (RFID) technology is rapidly popularized in the fields of intelligent transportation, logistics management, industrial automation, and the like, and has great development potential due to its fast and efficient data collection ability. RFID technology is widely used in the field of indoor localization, in which three-dimensional location can obtain more real and specific target location information. Aiming at the existing three-dimensional location scheme based on RFID, this paper proposes a new three-dimensional localization method based on deep learning: combining RFID absolute location with relative location, analyzing the variation characteristics of the received signal strength (RSSI) and Phase, further mining data characteristics by deep learning, and applying the method to the smart library scene. The experimental results show that the method has a higher location accuracy and better system stability.

show abstract

Section: Mining Datamentioning

confidence: 99%

Section: Mining Datamentioning

confidence: 99%

Section: Mining Datamentioning

confidence: 99%

See 1 more Smart Citation

3DLRA: An RFID 3D Indoor Localization Method Based on Deep Learning

Cheng

Wang

Guan

et al. 2020

Sensors

View full text Add to dashboard Cite

show abstract

“…Focusing on how to optimize the network performance, some scholars have carried out some relevant studies on 2DCNN. For example, Li et al [34] proposed a complex convolution kernel under the two-dimensional structure, and gave the suggestions on how to set the convolution kernels through comparative experiments. He et al [35] explored the relationship between network depth, width and size of convolution kernel through a large number of experiments to optimize the network structure.…”

Section: Introductionmentioning

confidence: 99%

Signal Status Recognition Based on 1DCNN and Its Feature Extraction Mechanism Analysis

Huang

Tang

Dai

et al. 2019

Sensors

View full text Add to dashboard Cite

In this paper, we construct a one-dimensional convolutional neural network (1DCNN), which directly takes as the input the vibration signal in the mechanical operation process. It can realize intelligent mechanical fault diagnosis and ensure the authenticity of signal samples. Moreover, due to the excellent interpretability of the 1DCNN, we can explain the feature extraction mechanism of convolution and the synergistic work ability of the convolution kernel by analyzing convolution kernels and their output results in the time-domain, frequency-domain. What’s more, we propose a novel network parameter-optimization method by matching the features of the convolution kernel with those of the original signal. A large number of experiments proved that, this optimization method improve the diagnostic accuracy and the operational efficiency greatly.

show abstract

“…This algorithm reduces the arithmetic complexity of the convolutional layer by using a minimal filtering technique. These approaches to compute the convolution can further be optimized by using different techniques and schemes [12][13][14].…”

Section: Introductionmentioning

confidence: 99%

Optimized Deep Neural Networks for Real-Time Object Classification on Embedded GPUs

2017

View full text Add to dashboard Cite

Abstract:Convolution is the most computationally intensive task of the Convolutional Neural Network (CNN). It requires a lot of memory storage and computational power. There are different approaches to compute the solution of convolution and reduce its computational complexity. In this paper, a matrix multiplication-based convolution (ConvMM) approach is fully parallelized using concurrent resources of GPU (Graphics Processing Unit) and optimized, considerably improving the performance of the image classifiers and making them applicable to real-time embedded applications. The flow of this CUDA (Compute Unified Device Architecture)-based scheme is optimized using unified memory and hardware-dependent acceleration of matrix multiplication. Proposed flow is evaluated on two different embedded platforms: first on an Nvidia Jetson TX1 embedded board and then on a Tegra K1 GPU of an Nvidia Shield K1 Tablet. The performance of this optimized and accelerated convolutional layer is compared with its sequential and heterogeneous versions. Results show that the proposed scheme significantly improves the overall results including energy efficiency, storage requirement and inference performance. In particular, the proposed scheme on embedded GPUs is hundreds of times faster than the sequential version and delivers tens of times higher performance than the heterogeneous approach.

show abstract

Complex convolution Kernel for deep networks

Cited by 4 publications

References 11 publications

3DLRA: An RFID 3D Indoor Localization Method Based on Deep Learning

3DLRA: An RFID 3D Indoor Localization Method Based on Deep Learning

Signal Status Recognition Based on 1DCNN and Its Feature Extraction Mechanism Analysis

Optimized Deep Neural Networks for Real-Time Object Classification on Embedded GPUs

Contact Info

Product

Resources

About