Method for optimizing a data model and device using the same

1. A device for optimizing the data model comprising:

a processor; and

a non-transitory storage medium coupled to the processor and configured to store a plurality of instructions, which cause the device to:

acquire data information and select at least two data models according to the data information;

utilize the data information to train the at least two data models;

acquire accuracies of each of the at least two data models, and determine a target data model which has a greatest accuracy between the at least two data models; and

optimize the target data model to obtain a first optimized data model;

determine whether an accuracy of the first optimized data model is greater than an accuracy of the target data model;

output the first data model if the accuracy of the first data model is greater than the accuracy of the target data model; and

optimize the first optimized data model to obtain a second optimized data model if the accuracy of the first data model is not greater than the accuracy of the target data model.

2. The device as recited in claim 1 , wherein the plurality of instructions are further configured to cause the device to:

reduce a quantity of neurons per layer in the target data model at a preset proportion.

3. The device as recited in claim 1 , wherein the plurality of instructions are further configured to cause the device to:

divide the data information into a training set and a verification set;

select a first data model from the least two data models, and utilize the training set to train the first data model;

verify the trained first data model by the verification set to obtain a first verification result, and determine an accuracy of the first data model according to the first verification result;

select a second data model from the at least two data models, and utilize the training set to train the second data model;

verify the trained second data model by the verification set to obtain a second verification result; and

determine an accuracy of the second data model according to the second verification result.

4. The device as recited in claim 1 , wherein the plurality of instructions are further configured to cause the device to:

obtain weights of every layer of neurons in the first optimized data model;

determine whether the weights of every layer of neurons are consistent with a gaussian distribution;

retain the layer of neurons if the weights of the layer of neurons are consistent with the gaussian distribution; and

adjust the layer of neurons if the weights of the layer of neurons are not consistent with the gaussian distribution.

5. The device as recited in claim 4 , wherein the plurality of instructions are further configured to cause the device to:

determine the layer of neurons which have weights not conforming to the Gaussian distribution;

obtain the weights corresponding to the determined layer of neurons;

sort the layer of neurons according to the weights; and

remove the neurons with a weight smaller than a preset weight at a preset removal ratio.

6. The device as recited in claim 5 , wherein the plurality of instructions are further configured to cause the device to:

acquire the validation set;

verify the second optimized data model by the validation set to obtain a verification result;

determine an accuracy of the second optimized data model according to the verification result;

determine whether the accuracy of the second optimized data model is greater than the accuracy of the target data model;

output the second optimized data model if the accuracy of the second optimized data model is greater than the accuracy of the target data model; and

optimize the second optimized data model to obtain a third optimized data model if the accuracy of the second optimized data model is not greater than the accuracy of the target data model.

7. A method for optimizing a data model comprising:

acquiring data information and selecting at least two data models according to the data information;

utilizing the data information to train the at least two data models;

acquiring each accuracy of the at least two data models, and determining a target data model which has greatest accuracy between the at least two data models; and

optimizing the target data model to obtain a first optimized data model;

determining whether an accuracy of the first optimized data model is greater than an accuracy of the target data model;

outputting the first data model if the accuracy of the first data model is greater than the accuracy of the target data model; and

optimizing the first optimized data model to obtain a second optimized data model if the accuracy of the first data model is not greater than the accuracy of the target data model.

8. The method as recited in claim 7 , further comprising:

reducing a quantity of neurons per layer in the target data model at a preset proportion.

9. The method as recited in claim 7 , further comprising:

dividing the data information into a training set and a verification set;

selecting a first data model of the at least two data models, and utilizing the training set to train the first data model;

verifying the trained first data model by the verification set to obtain a first verification result, and determining an accuracy of the first data model according to the first verification result;

selecting a second data model from the at least two data models, and utilizing the training set to train the second data model;

verifying the trained second data model by the verification set to obtain a second verification result; and

determining an accuracy of the second data model according to the second verification result.

10. The method as recited in claim 7 , further comprising:

obtaining weights of every layer of neurons in the first optimized data model;

determining whether the weights of every layer of neurons are consistent with a gaussian distribution;

retaining the layer of neurons if the weights of the layer of neurons are consistent with the gaussian distribution; and

adjusting the layer of neurons if the weights of the layer of neurons are not consistent with the gaussian distribution.

11. The method as recited in claim 10 , further comprising:

determining the layer of neurons which have weights not conforming to the Gaussian distribution;

obtaining the weights corresponding to the determined layer of neurons;

sorting the layer of neurons according to the weights; and

removing the neurons with a weight smaller than a preset weight at a preset removal ratio.

12. The method as recited in claim 11 , further comprising:

acquiring the validation set;

verifying the second optimized data model by the validation set to obtain a verification result;

determines an accuracy of the second optimized data model according to the verification result;

determining whether the accuracy of the second optimized data model is greater than the accuracy of the target data model;

outputting the second optimized data model if the accuracy of the second optimized data model is greater than the accuracy of the target data model; and

optimizing the second optimized data model to obtain a third optimized data model if the accuracy of the second optimized data model is not greater than the accuracy of the target data model.

13. A non-transitory storage medium having stored thereon instructions that, when executed by at least one processor of a device for optimizing a data model, causes the least one processor to execute instructions of a method for optimizing a data model, the method comprising:

acquiring data information and selecting at least two data models according to the data information;

utilizing the data information to train the at least two data models;

acquiring each accuracy of the at least two data models, and determining a target data model which has greatest accuracy between the at least two data models; and

optimizing the target data model to obtain a first optimized data model;

determining whether an accuracy of the first optimized data model is greater than an accuracy of the target data model;

outputting the first data model if the accuracy of the first data model is greater than the accuracy of the target data model; and

optimizing the first optimized data model to obtain a second optimized data model if the accuracy of the first data model is not greater than the accuracy of the target data model.

14. The non-transitory storage medium as recited in claim 13 , wherein the method is further comprising:

dividing the data information into a training set and a verification set;

selecting a first data model from the at least two data models, and utilize the training set to train the first data model;

verifying the trained first data model by the verification set to obtain a first verification result, and determine an accuracy of the first data model according to the first verification result;

selecting a second data model from the at least two data models, and utilizing the training set to train the second data model;

verifying the trained second data model by the verification set to obtain a second verification result; and

determining an accuracy of the second data model according to the second verification result.

15. The non-transitory storage medium as recited in claim 13 , wherein the method is further comprising:

obtaining weights of every layer of neurons in the first optimized data model;

determining whether the weights of every layer of neurons are consistent with a gaussian distribution;

retaining the layer of neurons if the weights of the layer of neurons are consistent with the gaussian distribution; and

adjusting the layer of neurons if the weights of the layer of neurons are not consistent with the gaussian distribution.

16. The non-transitory storage medium as recited in claim 15 , wherein the method is further comprising:

determining the layer of neurons which have weights not conforming to the gaussian distribution;

obtaining the weights corresponding to the determined layer of neurons;

sorting the layer of neurons according to the weights; and

removing the neurons with a weight smaller than a preset weight at a preset removal ratio.

17. The non-transitory storage medium as recited in claim 16 , wherein the method is further comprising:

acquiring the validation set;

verifying the second optimized data model by the validation set to obtain a verification result;

determining an accuracy of the second optimized data model according to the verification result;

determining whether the accuracy of the second optimized data model is greater than the accuracy of the target data model;

output the second optimized data model if the accuracy of the second optimized data model is greater than the accuracy of the target data model; and

optimizing the second optimized data model to obtain a third optimized data model if the accuracy of the second optimized data model is not greater than the accuracy of the target data model.