Distributed system for executing machine learning and machine learning execution method

Abstract

The invention provides a distributed system for executing machine learning and a machine learning execution method. The system comprises multiple computing devices and a parameter memory, wherein eachcomputing device is configured to execute data stream-oriented computation according to its own data record, and the data stream-oriented computation is expressed with one ore more directed acyclic graphs; the parameter memory is used for maintaining parameters of a machine learning model; when the data stream-oriented computation for training the machine learning model is executed, the computingdevices execute operation about machine learning model training according to respective data records by use of the parameters acquired from the parameter memory, and the parameter memory updates theparameters according to the operation results of the computing devices; and / or when the data stream-oriented computation for performing estimation by use of the machine learning model is executed, thecomputing devices execute operation about machine learning model estimation according to respective data records by use of the parameters acquired from the parameter memory. Therefore, operation overhead of machine learning can be reduced.

Description

technical field

[0001] Exemplary embodiments of the present invention generally relate to the field of artificial intelligence, and more particularly, to a distributed system for performing machine learning and a method for performing machine learning using the distributed system. Background technique

[0002] With the rapid growth of data scale, machine learning is widely used in various fields to mine the value of data. However, in order to perform machine learning on a large data scale, in practice, it is often necessary to use a distributed machine learning platform including multiple computing devices to complete the training of the machine learning model or the corresponding estimation.

[0003] In existing distributed machine learning systems (for example, Google's deep learning framework TensorFlow), if you want to implement multi-configuration or multiple runs based on a certain machine learning algorithm, or if you want to run multiple For machine learning algorit...

Images