Method of compressing neural network model and electronic apparatus for performing the same

Abstract

Disclosed is a method of compressing a neural network model that is performed by a computing device. The method includes receiving a trained model and compression method instructions for compressing the trained model, identifying a compressible block and a non-compressible block among a plurality of blocks included in the trained model based on the compression method instructions, transmitting a command to a user device that causes the user device to: display a structure of the trained model representing a connection relationship between the plurality of blocks on a first screen such that the compressible block and the non-compressible block are visually distinguished, and display, on a second screen, an input field operable to receive a parameter value entered by a user for compression of the compressible block, and compressing the trained model based on the parameter value entered by the user in the input field.

What is claimed is: 
     
         1 . A method of controlling server for compressing a neural network model including:
 receiving, at a processor of the server, a compression mode from a user device, the compression mode includes a first compression mode and a second compression mode;   controlling, via the processor, the user device to display a compression setting screen for receiving a configuring value, the compression setting screen is display differently according to the compression mode; and   controlling, via the processor, the user device to outputting download data in which the neural network model is compressed based on the configuring value,   wherein the neural network model includes a plurality of blocks,   wherein the configuring value of the first compression mode is a first configuring value for configuring a compression ratio for an entire neural network model,   wherein the configuring value of the second compression mode is a second configuring value for configuring a compression ratio for at least one block of the plurality of blocks.   
     
     
         2 . The method of  claim 1 , wherein the compression setting screen of the first compression mode includes,
 a base model input region for receiving a base model to be compressed, and   a configuring value input region for receiving the first configuring value for configuring the compression ratio for the base model.   
     
     
         3 . The method of  claim 1 , wherein the compression setting screen of the second compression mode includes,
 a block display region for displaying information on a first block which is at least one of the plurality of blocks being comprised a base model to be compressed, and   a configuring value input region for receiving the second configuring value for configuring the compression ratio for the first block.   
     
     
         4 . The method of  claim 3 , wherein the compression setting screen of the second compression mode includes an architecture display region for displaying an architecture indicating a connection relationship between the plurality of blocks of the base model,
 wherein the plurality of blocks of the base model includes a compressible block and non-compressible block, and   wherein the first block being displayed on the block display region is the compressible block.   
     
     
         5 . The method of  claim 3 , further including:
 controlling, via the processor, the user device to display a compression method setting screen prior to display the compression setting screen when the processor received the second compression mode,   wherein the compression method setting screen includes a base model input region for receiving the base model and a compression method setting region for receiving a compression method to be applied to the base model,   wherein at least a part of the information displayed on the compression setting screen of the second compression mode is different according to the compression method set in the compression method setting region.   
     
     
         6 . The method of  claim 1 , wherein the neural network model of the first compression mode is compressed using an individual parameter which is the compression ratio for at least one individual block of the plurality of blocks,
 wherein the individual parameter is obtained based on the first configuring value.   
     
     
         7 . The method of  claim 4 , wherein the compressible block and non-compressible block are displayed visually distinguished on the compression setting screen of the second compression mode. 
     
     
         8 . The method of  claim 7 , wherein the compressible block is selectable by user, and the non-compressible block is non-selectable by the user. 
     
     
         9 . The method of  claim 3 , wherein the second configuring value of the first block is calculated and input by the processor based on user input,
 wherein the processor controls the user device to be output the download data in which the neural network model is compressed based on a modified second configuration value when the modified second configuration is receive from the user device after the second configuration value of the first block is displayed via the user device.   
     
     
         10 . A server for compressing a neural network model including:
 a communication interface, configured to communicate with a user device, including at least one communication circuit; and   a processor connected to the communication interface,   wherein the processor configured to,   receiving, at a processor of the server, a compression mode from the user device, the compression mode includes a first compression mode and a second compression mode,   controlling, via the processor, the user device to display a compression setting screen for receiving a configuring value, the compression setting screen is display differently according to the compression mode, and   controlling, via the processor, the user device to outputting download data in which the neural network model is compressed based on the configuring value,   wherein the neural network model includes a plurality of blocks,   wherein the configuring value of the first compression mode is a first configuring value for configuring a compression ratio for an entire neural network model,   wherein the configuring value of the second compression mode is a second configuring value for configuring a compression ratio for at least one block of the plurality of blocks.   
     
     
         11 . The server of  claim 10 , wherein the compression setting screen of the first compression mode includes,
 a base model input region for receiving a base model to be compressed, and   a configuring value input region for receiving the first configuring value for configuring the compression ratio for the base model.   
     
     
         12 . The server of  claim 10 , wherein the compression setting screen of the second compression mode includes,
 a block display region for displaying information on a first block which is at least one of the plurality of blocks being comprised a base model to be compressed, and   a configuring value input region for receiving the second configuring value for configuring the compression ratio for the first block.   
     
     
         13 . The server of  claim 12 , wherein the processor further configured to,
 controlling, via the processor, the user device to display a compression method setting screen prior to display the compression setting screen when the processor received the second compression mode,   wherein the compression method setting screen includes a base model input region for receiving the base model and a compression method setting region for receiving a compression method to be applied to the base model,   wherein at least a part of the information displayed on the compression setting screen of the second compression mode is different according to the compression method set in the compression method setting region.   
     
     
         14 . The server of  claim 10 , wherein the neural network model of the first compression mode is compressed using an individual parameter which is the compression ratio for at least one individual block of the plurality of blocks,
 wherein the individual parameter is obtained based on the first configuring value.   
     
     
         15 . The server of  claim 12 , wherein the second configuring value of the first block is calculated and input by the processor based on user input,
 wherein the processor controls the user device to be output the download data in which the neural network model is compressed based on a modified second configuration value when the modified second configuration is receive from the user device after the second configuration value of the first block is displayed via the user device.   
     
     
         16 . A method of controlling a user device for compressing a neural network model including:
 receiving, at a processor of the user device, a compression mode from a user, the compression mode includes a first compression mode and a second compression mode;   displaying, via the processor, a compression setting screen for receiving a configuring value, the compression setting screen is display differently according to the compression mode; and   outputting, via the processor, download data in which the neural network model is compressed based on the configuring value,   wherein the neural network model includes a plurality of blocks,   wherein the configuring value of the first compression mode is a first configuring value for configuring a compression ratio for an entire neural network model,   wherein the configuring value of the second compression mode is a second configuring value for configuring a compression ratio for at least one block of the plurality of blocks.   
     
     
         17 . The method of  claim 16 , wherein the compression setting screen of the first compression mode includes,
 a base model input region for receiving a base model to be compressed, and   a configuring value input region for receiving the first configuring value for configuring the compression ratio for the base model.   
     
     
         18 . The method of  claim 16 , wherein the compression setting screen of the second compression mode includes,
 a block display region for displaying information on a first block which is at least one of the plurality of blocks being comprised a base model to be compressed, and   a configuring value input region for receiving the second configuring value for configuring the compression ratio for the first block.   
     
     
         19 . The method of  claim 18 , further including:
 displaying, via the processor, a compression method setting screen prior to display the compression setting screen when the processor received the second compression mode,   wherein the compression method setting screen includes a base model input region for receiving the base model and a compression method setting region for receiving a compression method to be applied to the base model,   wherein at least a part of the information displayed on the compression setting screen of the second compression mode is different according to the compression method set in the compression method setting region.   
     
     
         20 . The method of  claim 16 , wherein the neural network model of the first compression mode is compressed using an individual parameter which is the compression ratio for at least one individual block of the plurality of blocks,
 wherein the individual parameter is obtained based on the first configuring value.