Multidimensional data processing device, method, and computer-readable recording medium

1 . A multidimensional data processing device comprising:

a memory configured to store instructions; and

at least one processor configured to execute the instructions to:

generate by setting a number of elements of each dimension other than a predetermined one dimension to 1 based on multidimensional data corresponding to one input data, 1-dimensional data whose number of elements is the same as a number of elements of the predetermined one dimension in the multidimensional data;

reduce the number of elements included in the 1-dimensional data;

copy, multiple times, the 1-dimensional data whose number of elements has been reduced, thereby to generate second multidimensional data in which the number of elements of each dimension other than the predetermined one dimension is restored to original number of elements;

perform a convolution layer process with a filter size of 1×1 on the second multidimensional data, thereby to generate third multidimensional data in which the number of elements of the predetermined one dimension in the second multidimensional data is restored to original number of elements; and

perform an element-wise product operation, based on the multidimensional data corresponding to one input data and the third multidimensional data,

wherein a number of elements of the 1-dimensional data whose number of elements has been reduced and the number of elements of the predetermined one dimension in the second multidimensional data are the same, and

wherein when A denotes the number of elements of the 1-dimensional data whose number of elements has been reduced, i denotes an integer from 0 to A-1, each i-th element of the predetermined one dimension in the second multidimensional data is the same as i-th element of the predetermined one dimension in the 1-dimensional data whose number of elements has been reduced.

2 . The multidimensional data processing device according to claim 1 ,

wherein the at least one processor is configured to apply a sigmoid function to each element in the third multidimensional data;

wherein the at least one processor is configured to perform the element-wise product operation of the multidimensional data corresponding to one input data and the third multidimensional data after the sigmoid function is applied to each element.

3 . The multidimensional data processing device according to claim 1 ,

wherein the at least one processor is configured to change values of the elements in the 1-dimensional data whose number of elements has been reduced that have negative values to 0;

wherein the at least one processor is configured to copy the 1-dimensional data multiple times.

4 . The multidimensional data processing device according to claim 1 ,

wherein the multidimensional data corresponding to one input data is 3 dimensional data, and,

the predetermined one dimension is a dimension of channel.

5 . A multidimensional data processing method comprising:

generating, by setting a number of elements of each dimension other than a predetermined one dimension to 1 based on multidimensional data corresponding to one input data, 1-dimensional data whose number of elements is the same as a number of elements of the predetermined one dimension in the multidimensional data;

reducing the number of elements included in the 1-dimensional data;

executing a copy process of copying, multiple times, the 1-dimensional data whose number of elements has been reduced, thereby to generate second multidimensional data in which the number of elements of each dimension other than the predetermined one dimension is restored to original number of elements;

performing a convolution layer process with a filter size of 1×1 on the second multidimensional data generated by the copy process, thereby to generate third multidimensional data in which the number of elements of the predetermined one dimension in the second multidimensional data generated by the copy process is restored to original number of elements; and

performing an element-wise product operation, based on the multidimensional data corresponding to one input data and the third multidimensional data whose number of elements in the predetermined one dimension is restored to original number of elements,

wherein a number of elements of the 1-dimensional data whose number of elements has been reduced and the number of elements of the predetermined one dimension in the second multidimensional data are the same, and

wherein when A denotes the number of elements of the 1-dimensional data whose number of elements has been reduced, i denotes an integer from 0 to A-1, each i-th element of the predetermined one dimension in the second multidimensional data is the same as i-th element of the predetermined one dimension in the 1-dimensional data whose number of elements has been reduced.

6 . The multidimensional data processing method according to claim 5 , further comprising:

applying a sigmoid function to each element in the third multidimensional data whose number of elements of the predetermined one dimension is restored to original number of elements,

wherein performing the element-wise product operation includes performing the element-wise product operation of the multidimensional data corresponding to one input data and the third multidimensional data after the sigmoid function is applied to each element.

7 . A non-transitory computer-readable recording medium in which a multidimensional data processing program is recorded, wherein the multidimensional data processing program causes a computer to execute:

a 1-dimensional data generation process of generating, by setting a number of elements of each dimension other than a predetermined one dimension to 1 based on multidimensional data corresponding to one input data, 1-dimensional data whose number of elements is the same as a number of elements of the predetermined one dimension in the multidimensional data;

a number of elements reducing process of reducing the number of elements included in the 1-dimensional data;

a copy process of copying, multiple times, the 1-dimensional data whose number of elements has been reduced, thereby to generate second multidimensional data in which the number of elements of each dimension other than the predetermined one dimension is restored to original number of elements;

a multidimensional data generation process of performing a convolution layer process with a filter size of 1×1 on the second multidimensional data generated by the copy process, thereby to generate third multidimensional data in which the number of elements of the predetermined one dimension in the second multidimensional data generated by the copy process is restored to original number of elements; and

an element-wise product operation process of performing an element-wise product operation, based on the multidimensional data corresponding to one input data and the third multidimensional data generated by the multidimensional data generation process,

wherein a number of elements of the 1-dimensional data whose number of elements has been reduced and the number of elements of the predetermined one dimension in the second multidimensional data are the same, and

wherein when A denotes the number of elements of the 1-dimensional data whose number of elements has been reduced, i denotes an integer from 0 to A-1, each i-th element of the predetermined one dimension in the second multidimensional data is the same as i-th element of the predetermined one dimension in the 1-dimensional data whose number of elements has been reduced.

8 . The non-transitory computer-readable recording medium in which the multidimensional data processing program is recorded according to claim 7 ,

wherein the multidimensional data processing program causes the computer to execute:

a sigmoid function process of applying a sigmoid function to each element in the third multidimensional data generated by the multidimensional data generation process,

wherein the multidimensional data processing program causes the computer to execute in the element-wise product operation process,

performing the element-wise product operation of the multidimensional data corresponding to one input data and the third multidimensional data after the sigmoid function is applied to each element.