Method, electronic device and computer program product for processing machine learning model

Patent metadata

Field	Value
Publication number	US-2021034582-A1
Application number	US-201916678758-A
Country	US
Kind code	A1
Filing date	Nov 8, 2019
Priority date	Aug 2, 2019
Publication date	Feb 4, 2021
Grant date	—

Abstract

A method comprises obtaining an intermediate representation of the machine learning model written in a source language. The intermediate representation is independent of the source language and a target language and comprises a structured text. The method also comprises generating a computation graph based on the intermediate representation. Nodes in the computation graph represent functions related to the machine learning model and a directed edge in the computation graph represents a dependency between functions. The method further comprises partitioning the computation graph into sequential parts in sequence such that the parts are executed sequentially, and functions corresponding to nodes in each part are executed in parallel.

First claim

What is claimed is:

1 . A method for processing a machine learning model, comprising:
obtaining an intermediate representation of the machine learning model written in a source language, the intermediate representation being independent of the source language and a target language and comprising a structured text; generating a computation graph based on the intermediate representation, nodes in the computation graph representing functions related to the machine learning model, a directed edge in the computation graph representing a dependency between the functions; and partitioning the computation graph into sequential parts, such that the parts are executed sequentially and functions corresponding to nodes in each part are executed in parallel.

2 . The method of claim 1 , wherein partitioning the computation graph into parts comprises determining in-degrees of at least a part of the plurality of nodes of the computation graph, an in-degree of a node representing a number of directed edges directed to the node.

3 . The method of claim 2 , wherein partitioning the computation graph into parts comprises partitioning the computation graph into the parts based on the in-degrees.

4 . The method of claim 3 , wherein partitioning the computation graph into the parts based on the in-degrees comprises executing iteratively acts of selecting a first part of the computation graph, such that each node in the first part has a predetermined threshold in-degree.

5 . The method of claim 4 , wherein partitioning the computation graph into the parts based on the in-degrees comprises executing iteratively acts of removing, from the computation graph, the first part and a directed edge associated with nodes in the first part, to update the computation graph.

6 . The method of claim 4 , wherein the threshold in-degree is equal to zero.

7 . An electronic device, comprising:
a processor; and a memory storing computer program instructions, the processor running the computer program instructions in the memory to control the electronic device to perform acts comprising:
obtaining an intermediate representation of the machine learning model written in a source language, the intermediate representation being independent of the source language and a target language and comprising a structured text;
generating a computation graph based on the intermediate representation, nodes in the computation graph representing functions related to the machine learning model, a directed edge in the computation graph representing a dependency between the functions; and
partitioning the computation graph into sequential parts in sequence, such that the plurality of parts are executed sequentially, and functions corresponding to nodes in each part are executed in parallel.

8 . The electronic device of claim 7 , wherein partitioning the computation graph into parts comprises determining in-degrees of at least a part of nodes in the plurality of nodes of the computation graph, an in-degree of a node representing a number of directed edges directed to the node.

9 . The electronic device of claim 8 , wherein partitioning the computation graph into parts comprises partitioning the computation graph into the parts based on the in-degrees.

10 . The electronic device of claim 9 , wherein partitioning the computation graph into the plurality of parts based on the in-degrees comprises executing iteratively acts of selecting a first part of the computation graph, such that each node in the first part has a predetermined threshold in-degree.

11 . The electronic device of claim 10 , wherein partitioning the computation graph into the plurality of parts based on the in-degrees comprises executing iteratively acts of removing, from the computation graph, the first part and a directed edge associated with nodes in the first part, to update the computation graph.

12 . The electronic device of claim 10 , wherein the threshold in-degree is equal to zero.

13 . A computer program product, being tangibly stored on a non-volatile computer-readable medium and comprising machine executable instructions, the machine executable instructions, when executed, causing a machine to perform acts comprising:
obtaining an intermediate representation of the machine learning model written in a source language, the intermediate representation being independent of the source language and s target language and including a structured text; generating a computation graph based on the intermediate representation, a node in the computation graph representing functions related to the machine learning model, a directed edge in the computation graph representing a dependency between the functions; and partitioning the computation graph into sequential parts, such that parts are executed sequentially, and functions corresponding to nodes in each part are executed in parallel.

14 . The computer program product of claim 13 , wherein partitioning the computation graph into parts comprises determining in-degrees of at least a part of nodes in the plurality of nodes of the computation graph, an in-degree of a node representing a number of a directed edge directed to the node.

15 . The computer program product of claim 14 , wherein partitioning the computation graph into parts comprises partitioning the computation graph into parts based on the in-degrees based on the in-degrees.

16 . The computer program product of claim 15 , wherein partitioning the computation graph into parts based on the in-degrees comprises executing iteratively acts of selecting a first part of the computation graph, such that each node in the first part has a predetermined threshold in-degree.

17 . The computer program product of claim 16 , wherein partitioning the computation graph into parts based on the in-degrees comprises executing iteratively acts of removing, from the computation graph, the first part and directed edges associated with nodes in the first part and the first part, to update the computation graph.

18 . The computer program product of claim 16 , wherein the threshold in-degree is equal to zero.

19 . The computer program product of claim 13 , wherein the intermediate representation of the machine learning model is described in a JavaScript Object Notation (JSON) format.

20 . The computer program product of claim 13 , wherein the intermediate representation of the machine learning model is described in an Extensible Markup Language (XML) format.

Assignees

• Emc Ip Holding Co Llc

Inventors

• Liu Jinpeng
• Wu Pengfei
• Ying Zhi
• Zhao Min

Classifications

• G06N3/045

  Combinations of networks · CPC title

• G06F8/41

  Compilation · CPC title

• G06N3/08

  Learning methods · CPC title

• G06F8/30 Primary

  Creation or generation of source code · CPC title

• G06N20/00 Primary

  Machine learning · CPC title