Method and electronic device for generating optimal neural network (NN) model

An electronic device for generating a neural network (NN) model, wherein the electronic device comprises:

• at least one memory storing instructions;
• and
• at least one processor configured to execute the instructions to:
• receive an image and a request for performing facial recognition;
• determine whether the image comprises the facial image
• identify a set of layers, which produces an output whose difference from a final output of the NN model is less than a preset threshold, among a plurality of layers of the NN model;
• identify a plurality of intermediate exit gates and a location of each of the plurality of intermediate exit gates within the set of layers;
• obtain intermediate outputs of the NN model by passing an input dataset through each intermediate exit gate of the plurality of intermediate exit gates;
• obtain an accuracy score for each of the plurality of intermediate exit gates based on a comparison of the final output of the NN model with each of the intermediate outputs;
• identify an earliest intermediate exit gate whose accuracy score is within a preset accuracy score based on the location of each of the plurality of intermediate exit gates within the set of layers, among the plurality of intermediate exit gates, and based on a determination of whether the image comprises the facial image, wherein the earliest intermediate exit gate is closest to an input layer of the NN model among at least one intermediate exit gate whose accuracy score is within the preset accuracy score;
• and
• generate the NN model by removing or dropping remaining layers of the plurality of layers and remaining intermediate exit gates of the plurality of intermediate exit gates located after the determined earliest intermediate exit gate.

A non-transitory computer-readable storage medium storing a program that is executable by at least one processor to perform a method for generating a neural network (NN) model, the method comprising:

• receiving an image and a request for performing facial recognition;
• determining whether the image comprises the facial image;
• identifying a set of layers, which produces an output whose difference from a final output of the NN model is less than a preset threshold, among a plurality of layers of the NN model;
• identifying a plurality of intermediate exit gates and a location of each of the plurality of intermediate exit gates within the set of layers;
• obtaining intermediate outputs of the NN model by passing an input dataset through each of the plurality of intermediate exit gates;
• obtaining an accuracy score for each of the plurality of intermediate exit gates based on a comparison of the final output of the NN model with each of the intermediate outputs;
• identifying an earliest intermediate exit gate whose accuracy score is within a preset accuracy score based on the location of each of the plurality of intermediate exit gates within the set of layers, among the plurality of intermediate exit gates, and based on a determination of whether the image comprises the facial image, wherein the earliest intermediate exit gate is closest to an input layer of the NN model among at least one intermediate exit gate whose accuracy score is within the preset accuracy score;
• and
• generating the NN model by removing or dropping remaining layers of the plurality of layers and remaining intermediate exit gates of the plurality of intermediate exit gates located after the determined earliest intermediate exit gate.