Machine learning program, optimization program, machine learning method, optimization method, and information processing device.

Abstract

To provide an information processor, a machine learning method, and a program, and an optimization method that execute machine learning using a frequency image with high accuracy.SOLUTION: In an information processor, a machine learning execution unit acquires a projection image in certain projection directions θ, φ from a projection image data table, generates a first frequency image by performing a Fourier transform on the projection image, inputs the first frequency image to an encoder, estimates a three-dimensional density structure in a Fourier space from a decoder, and acquires a second frequency image by projecting the three-dimensional density structure in projection directions θ and φ, sets a difference between the first frequency image and the second frequency image to Pθ,φ(u',v'), calculates, on the basis of the following formula (3), an estimation error, and executes machine learning of an Auto-Encoder type neural network to reduce the estimation error.SELECTED DRAWING: Figure 2

Claims

1. By inputting the output of the encoder, which has the first frequency image as input, into the decoder, a second frequency image is obtained. Training of the encoder and decoder is performed based on a loss function in which the weights for a first frequency are smaller than the weights for a second frequency that is higher than the first frequency, and on the first frequency image and the second frequency image. A machine learning program that instructs a computer to perform a task.

2. The machine learning program according to claim 1, wherein the loss function is a function that calculates an estimation error by accumulating values ​​obtained by multiplying the difference between the first frequency image and the second frequency image at each frequency coordinate by a weight for the first frequency which is higher than the weight for the second frequency which is lower, and the process of training the encoder and the decoder is to train the encoder and the decoder based on the estimation error.

3. The machine learning program according to claim 1, characterized in that the acquisition process involves inputting the output of the encoder, which has received the first frequency image as input, to the decoder to estimate the three-dimensional density structure, and acquiring the second frequency image based on the three-dimensional density structure.

4. The machine learning program according to claim 2, characterized in that the process of training the encoder and the decoder further executes a process of setting the difference between the first frequency image and the second frequency image at a certain frequency coordinate to 0 if the value based on a certain frequency coordinate of the first frequency image or the second frequency image is greater than a threshold.

5. The machine learning program according to claim 2, wherein the loss function further includes a Gaussian filter, and the process of training the encoder and the decoder is characterized in that the encoder and the decoder are trained based on an estimation error calculated based on the loss function further including the Gaussian filter.

6. The difference between a first frequency image based on a projection image obtained by projecting the first three-dimensional density structure in a certain projection direction and a second frequency image obtained by projecting the second three-dimensional density structure of Fourier space in the same projection direction is multiplied by a weight for the first frequency that is lower than the weight for the second frequency, where the weight for the first frequency is higher than the weight for the first frequency. The values ​​of the first frequency image are adjusted so that the multiplication result is small. An optimization program that causes a computer to perform a process.

7. By inputting the output of the encoder, which has the first frequency image as input, into the decoder, a second frequency image is obtained. Training of the encoder and decoder is performed based on a loss function in which the weights for a first frequency are smaller than the weights for a second frequency that is higher than the first frequency, and on the first frequency image and the second frequency image. A machine learning method in which a computer performs a process.

8. The difference between a first frequency image based on a projection image obtained by projecting the first three-dimensional density structure in a certain projection direction and a second frequency image obtained by projecting the second three-dimensional density structure of Fourier space in the same projection direction is multiplied by a weight for the first frequency that is lower than the weight for the second frequency, where the weight for the first frequency is higher than the weight for the first frequency. The values ​​of the first frequency image are adjusted so that the multiplication result is small. An optimization method used by computers to perform processing.

9. By inputting the output of the encoder, which has the first frequency image as input, into the decoder, a second frequency image is obtained. Training of the encoder and decoder is performed based on a loss function in which the weights for a first frequency are smaller than the weights for a second frequency that is higher than the first frequency, and on the first frequency image and the second frequency image. An information processing device having a control unit that performs processing.

10. The difference between a first frequency image based on a projection image obtained by projecting the first three-dimensional density structure in a certain projection direction and a second frequency image obtained by projecting the second three-dimensional density structure of Fourier space in the same projection direction is multiplied by a weight for the first frequency that is lower than the weight for the second frequency, where the weight for the first frequency is higher than the weight for the first frequency. The values ​​of the first frequency image are adjusted so that the multiplication result is small. An information processing device having a control unit that performs processing.

Images