Acceleration method for exploring optimization space in deep learning compiler

A technology for optimizing space and deep learning, applied in neural learning methods, compiler construction, parser generation, etc., can solve the problems of large optimization space and huge time-consuming exploration of operator optimization space, so as to reduce time consumption and save Overhead, time-consuming effects

Active Publication Date: 2021-03-30
ZHEJIANG LAB
View PDF3 Cites 8 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

The optimization space of each operator is very large. For example, a Conv operator may have hundreds of millions of optimization schemes. Therefore, it takes a lot of time to explore the optimization space of operators. For example, a Yolo network needs more than one day to explore and optimize. Program

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
  • Acceleration method for exploring optimization space in deep learning compiler
  • Acceleration method for exploring optimization space in deep learning compiler
  • Acceleration method for exploring optimization space in deep learning compiler

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0045] Specific embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings. It should be understood that the specific embodiments described here are only used to illustrate and explain the present invention, and are not intended to limit the present invention.

[0046] Such as figure 1As shown in , an acceleration method for exploring the optimization space in deep learning compilers, the purpose is to greatly reduce the time spent by the compiler in exploring the optimization space of operators at the expense of an acceptable increase in the inference time of deep learning networks. This method first abstracts the neural network into the form of a computational graph. Secondly, graph optimization is performed on the calculation graph, and an optimization space is defined for each operator in the optimized calculation graph. Then, based on the operator containing the optimal spatial information, a calculation method f...

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 invention discloses an acceleration method for exploring an optimization space in a deep learning compiler, and aims to optimize the effect of a neural network through a compiling technology and greatly reduce the time consumed for exploring an operator optimization space by the compiler. The method comprises the steps of firstly abstracting a neural network into a form of a calculation graph;secondly, performing graph optimization on the calculation graph, and defining an optimization space for each operator in the optimized calculation graph; and then, based on an operator containing optimization space information, providing an optimization space similarity calculation method. Finally, an operator state space exploration method based on similarity is provided, operators are clustered based on similarity, full-space exploration is carried out on a core operator in each cluster, other operators of the same type in an optimal scheme of the core operator are explored, and an optimization scheme of each operator of the whole neural network is determined.

Description

technical field [0001] The invention relates to the application fields of deep learning, compiling technology, and high-performance computing intersecting technology, and in particular to an acceleration method for exploring optimization space in a deep learning compiler. Background technique [0002] Today, deep neural networks (DNNs) have been widely used in image classification, natural language processing, autonomous driving, augmented reality, and other AI fields. Especially with the rapid development of computing equipment, such as the emergence of GPU, FPGA and specially designed neural network accelerators, the computing power of DNN is becoming more and more powerful, and the demand for efficient DNN in the field of artificial intelligence is also becoming stronger, so how to improve the operating efficiency of DNN It is a very important research problem in recent years. [0003] Now, there are already many deep learning frameworks, such as TensorFlow, PyTorch, Caf...

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
IPC IPC(8): G06F8/30G06N3/04G06N3/08
CPCG06F8/37G06N3/08G06N3/045
Inventor 潘秋红何水兵陈刚杨弢
Owner ZHEJIANG LAB
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap