Patent
Method and device for adapting to feature data in convolution neural network
Li Jianjun,Huang Chang,Chen Liang,Ling Kun,Li Delin +4 more
- 13 Mar 2018
4
TL;DR: In this paper, a method and device for adapting to the feature data in a convolution neural network is presented. But the method comprises the steps: selecting a plurality of continuous layers, determining an expected number of subdata blocks of output feature data of the last layer and the layout position, width and height of each subdata block; determining the layout positions, width, and height for each sub data block of each layer.
read more
Abstract: The invention discloses a method and device for adapting to the feature data in a convolution neural network. The method comprises the steps: selecting a plurality of continuous layers; determining anexpected number of subdata blocks of output feature data of the last layer and the layout position, width and height of each subdata block; determining the layout position, width and height of each subdata block of input feature data of each layer; determining the actual position of each subdata block of input feature data of the first layer; and obtaining an expected number of pieces of input subdata of the input feature data of the first layer accordignto the actual position of each subdata block of input feature data of the first layer. The method can achieve the efficient implementation of the convolution operation of the feature data at any size while effectively reducing the data transferred between an external storage unit and a processor.
read more
Chat with Paper
AI Agents for this Paper
Find similar papers on Google Scholar, PubMed and Arxiv
Write a critical review of this paper
Analyze citations of this paper to find unaddressed research gaps
Citations
Patent
A data reading method for a multi-core image processing convolutional neural network
Liu Mingrun
- 14 Jun 2019
TL;DR: In this paper, a data reading method for a multi-core image processing convolutional neural network is proposed, and the method comprises the steps: A), receiving convolution calculation parameters whichcomprise an image size, a channel number, a convolution kernel size KxK, and/or a step length; B) determining the number of segmentation times of each line of data of the image and the length of the last remaining data, so that the segmented image blocks generate M times of convolution required data each time; C) performing image segmentation according
2
Patent
Neural network data preprocessing and postprocessing method and device
TL;DR: In this paper, a neural network data preprocessing and postprocessing method and device is described, which comprises the steps that according to data postprocessing, an output channelset sequence of a designated layer of neural network is determined; and according to the output channel set sequence and design parameters of the designated layer, an operation parameter sequence ofthe designated layer is generated.
1
Patent
Feature data extraction method and apparatus and instruction generation method and device
Li Zhichao,Wang Zhenjiang,Ling Kun,Li Jianjun +3 more
- 15 Oct 2019
TL;DR: In this paper, a feature data extraction method and an apparatus and an instruction generation method and a device are described. But the method does not consider the second output feature data in the first convolution kernel.
Patent
Processing for multiple input data sets
Dana M. Vantrease,Ron Diamant,Volpe Thomas A,Huang Richard +3 more
- 26 Sep 2019
TL;DR: In this article, the authors present techniques for performing multi-layer neural network processing for multiple contexts, where a computing engine is set in a first configuration to implement a second layer of a neural network and to process first data related to a first context to generate first context second layer output.
References
Patent
Apparatus and Method for Achieving Accelerator of Sparse Convolutional Neural Network
Xie Dongliang,Zhang Yu,Shan Yi +2 more
- 10 Oct 2017
TL;DR: In this paper, the authors proposed an apparatus and method for realizing an accelerator of a sparse convolutional neural network, which consists of a CNN, a pooling unit, a full connection unit and a control unit.
49
Patent
Memory efficiency for convolutional neural networks operating on graphics processing units
Yi Yang,Chao Li,Min Feng,Srimat Chakradhar +3 more
- 20 May 2016
TL;DR: In this paper, the authors examine the performance impact of different data layouts and then describe a method to produce data layout selection for various layers of the CNN, including a fast transformation implementation.
36
Patent
Data-optimized neural network traversal
Joo-Hoon Lee
- 26 May 2016
TL;DR: In this article, the output tile of a second layer of a neural network can be generated using the processor by processing the output tiles of the first layer stored in the internal memory of the processor.
30
Patent
Calculating method and apparatus in convolution neural network
TL;DR: In this paper, the authors propose a calculating method and apparatus in a convolution neural network, which includes the steps: receiving original data; according to distribution of the original data, determining a first optimal quantified step, according to the first quantification step, performing fixed-point processing on the original dataset to generate first data; inputting the first data to the second layer of the convolutional neural network to generate the second dataset.
6
Patent
Parallel development and deployment for machine learning models
Qin Bin,Azam Farooq,Malov Denis +2 more
- 09 Jun 2016
TL;DR: In this article, the authors present a learning model for a data set where a subset of the inputs is selected, and the sample data set is partitioned into a number of partitions equal to a combined number of states of the selected inputs.
5
Related Papers (5)
Yuwei Hu,Jiangming Jin,Su Lei,Dinghua Li +3 more
- 21 Jun 2018
Dai Yan,Li Yanrong,Yao Da +2 more
- 06 Nov 2018
Han Dong Il,Jo Joo Yeon +1 more
- 09 May 2019