Unlock instant, AI-driven research and patent intelligence for your innovation.

Deep learning network processing method, device and compiler

A deep learning network and deep learning technology, applied in the field of devices, compilers, and deep learning network processing methods, can solve problems such as limited processing performance, and achieve the effect of improving performance and reducing memory access

Active Publication Date: 2021-11-30
SHENZHEN INTELLIFUSION TECHNOLOGIES CO LTD
View PDF0 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004] However, based on the current compilation technology, the deep learning processor and the on-chip storage unit still need to perform frequent memory access, so the processing performance is limited

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Deep learning network processing method, device and compiler
  • Deep learning network processing method, device and compiler
  • Deep learning network processing method, device and compiler

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0027] In order to facilitate the understanding of the technical solutions in the embodiments of the present application, some of the terms involved in the embodiments of the present application are firstly explained below:

[0028] op: operation, operation operation. An independent operation in the deep learning network, such as: convolution operation, pooling operation and activation operation, etc.; a network layer (layer) of the deep learning network contains at least one op.

[0029] LG: Layer Group, network layer grouping; obtained by splitting the deep learning network, where each LG includes multiple consecutive ops. In specific splitting, one LG may only contain one op. In the embodiment of this application, the processing is performed on LGs containing multiple ops.

[0030] LG segmentation technology: segment the feature map of LG. Among them, the segmentation operation process for a single op is as follows:

[0031] The segmentation of op is actually the segment...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The application provides a deep learning network processing method, device and compiler, which relate to the field of deep learning technology, wherein the method includes: splitting the deep learning network into at least one network layer grouping, wherein each network layer grouping contains Multiple continuous operations; for each network layer grouping, segment the network layer grouping to obtain the fragmentation of the network layer grouping; for each fragmentation of the network layer grouping, each operation related to the Operations are grouped together and compiled to generate object code blocks. The technical solution provided by this application can realize the segmentation of the network layer grouping and the operation combination process of the fragmentation in the compiler, and effectively improve the performance of the deep learning processor.

Description

technical field [0001] The present application relates to the technical field of deep learning, in particular to a deep learning network processing method, device and compiler. Background technique [0002] In recent years, deep learning networks have been widely used in various fields, and there are more and more deep learning frameworks and terminals based on deep learning networks. Among them, the current mainstream deep learning frameworks include TensorFlow, MXNet, Keras, and PyTorch. Programmable GateArray, FPGA) and other general-purpose deep learning processors and other special-purpose deep learning processors, etc., these deep learning frameworks and terminals are different, obviously, implementing different deep learning frameworks in a point-to-point manner It is unrealistic to provide back-end support for different terminals, so a compiler that supports various front-end deep learning frameworks and back-end terminals is particularly important. [0003] Based ...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Patents(China)
IPC IPC(8): G06N3/063G06N3/08
CPCG06N3/063G06N3/08
Inventor 王东
Owner SHENZHEN INTELLIFUSION TECHNOLOGIES CO LTD