公开(公告)号：US20200302611A1

公开(公告)日：2020-09-24

申请号：US16811108

申请日：2020-03-06

Applicant: FUJITSU LIMITED

Inventor： Kento UEMURA ,  Suguru YASUTOMI ,  TAKASHI KATOH

IPC: G06T7/11 ,  G06K9/62 ,  G06N3/08

Abstract: An estimation method implemented by a computer, the estimation method includes: executing learning processing by training an autoencoder with a data group corresponding to a specific task; calculating a degree of compression of each part regarding data included in the data group by using the trained autoencoder; and estimating a common part with another piece of data included in the data group regarding the data corresponding to the specific task based on the calculated degree of compression of each part.

公开(公告)号：US20200234081A1

公开(公告)日：2020-07-23

申请号：US16736911

申请日：2020-01-08

Applicant: FUJITSU LIMITED

Inventor： TAKASHI KATOH ,  Kento UEMURA ,  Suguru YASUTOMI

IPC: G06K9/62

Abstract: A learning device learns at last one parameter of a learning model such that each intermediate feature quantity becomes similar to a reference feature quantity, the each intermediate feature quantity being calculated as a result of inputting a plurality of sets of augmentation training data to a first neural network in the learning model, the plurality of augmentation training data being generated by performing data augmentation based on same first original training data. The learning device learns at last one parameter of a second network, in the learning model, using second original training data, which is different than the first original training data, and using the reference feature quantity.

公开(公告)号：US20190286946A1

公开(公告)日：2019-09-19

申请号：US16274321

申请日：2019-02-13

Applicant: FUJITSU LIMITED

Inventor： Kento UEMURA ,  TAKASHI KATOH ,  Suguru YASUTOMI ,  Toshio Endoh

IPC: G06K9/62 ,  G06N3/08

Abstract: A learning method for an auto-encoder is performed by a computer. The method includes: by using a discriminator configured to generate an estimated label based on a feature value generated by an encoder of an auto-encoder and input data, causing the discriminator to learn such that a label corresponding the input data and the estimated label are matched; and by using the discriminator, causing the encoder to learn such that the label corresponding to the input data and the estimated label are separated.

公开(公告)号：US20230289659A1

公开(公告)日：2023-09-14

申请号：US18174625

申请日：2023-02-26

Applicant: Fujitsu Limited

Inventor： Takashi KATOH ,  Kento UEMURA ,  Suguru YASUTOMI

IPC: G06N20/00

CPC classification number: G06N20/00

Abstract: An information processing method comprising: for a classification model for classifying input data into one or another of plural classes that was trained using a first data set, identifying, in a second data set that is different from the first data set one or more items of data having a specific datum of which a degree of contribution to a change in a classification criterion is greater than a predetermined threshold, the classification criterion being a classification criterion of the classification model during re-training based on the second data set; and, from among the one or more items of data, detecting an item of data, for which a loss reduces for the classification model by change to the classification criterion by re-training based on the second data set, as an item of data of an unknown class not contained in the plural classes.

公开(公告)号：US20210012193A1

公开(公告)日：2021-01-14

申请号：US16921944

申请日：2020-07-07

Applicant: FUJITSU LIMITED

Inventor： Suguru YASUTOMI ,  TAKASHI KATOH ,  Kento UEMURA

IPC: G06N3/08 ,  G06N3/04

Abstract: A machine learning method includes: calculating, by a computer, a first loss function based on a first distribution and a previously set second distribution, the first distribution being a distribution of a feature amount output from an intermediate layer when first data is input to an input layer of a model that has the input layer, the intermediate layer, and an output layer; calculating a second loss function based on second data and correct data corresponding to the first data, the second data being output from the output layer when the first data is input to the input layer of the model; and training the model based on both the first loss function and the second loss function.

公开(公告)号：US20200242412A1

公开(公告)日：2020-07-30

申请号：US16774100

申请日：2020-01-28

Applicant: FUJITSU LIMITED

Inventor： TAKASHI KATOH ,  Kazuki IWAMOTO ,  Kento UEMURA ,  Suguru YASUTOMI

IPC: G06K9/62 ,  G06T7/00 ,  G06N20/00 ,  G16H30/40

Abstract: An anomaly detection apparatus performs training for the generator and the discriminator such that the generator maximizes a discrimination error of the discriminator and the discriminator minimizes the discrimination error The anomaly detection apparatus stores, while the training is being performed, a state of the generator that is half-trained and satisfies a pre-set condition, and retrains the discriminator by using an image generated by the half-trained generator that has the stored state.

公开(公告)号：US20200234140A1

公开(公告)日：2020-07-23

申请号：US16741860

申请日：2020-01-14

Applicant: FUJITSU LIMITED

Inventor： TAKASHI KATOH ,  Kento UEMURA ,  Suguru YASUTOMI ,  Takeshi OSOEKAWA

IPC: G06N3/08 ,  G06N3/04 ,  G06N20/20

Abstract: A learning method executed by a computer, the learning method includes: learning parameters of a machine learning model having intermediate feature values by inputting a plurality of augmented training data, which is generated by augmenting original training data, to the machine learning model so that specific intermediate feature values, which are calculated from specific augmented training data augmented from a same original training data, become similar to each other.

公开(公告)号：US20200234139A1

公开(公告)日：2020-07-23

申请号：US16741839

申请日：2020-01-14

Applicant: FUJITSU LIMITED

Inventor： TAKASHI KATOH ,  Kento UEMURA ,  Suguru YASUTOMI

IPC: G06N3/08 ,  G06N20/20 ,  G06N3/04

Abstract: A learning method executed by a computer, the learning method including augmenting original training data based on non-stored target information included in the original training data to generate a plurality of augmented training data, generating a plurality of intermediate feature values by inputting the plurality of augmented training data to a learning model, and learning a parameter of the learning model such that, with regard to the plurality of intermediate feature values, each of the plurality of intermediate feature values generated from a plurality of augmented training data, augmented from reference training data, becomes similar to a reference feature value.

公开(公告)号：US20200226494A1

公开(公告)日：2020-07-16

申请号：US16717563

申请日：2019-12-17

Applicant: FUJITSU LIMITED

Inventor： Suguru YASUTOMI ,  Kento UEMURA ,  TAKASHI KATOH

IPC: G06N20/00 ,  G06T5/50 ,  G06F17/18 ,  A61B8/08

Abstract: A non-transitory computer-readable recording medium stores therein a learning program that causes a computer to execute a process including: generating a shadow image including a shadow according to a state of ultrasound reflection in an ultrasound image; generating a combined image by combining the ultrasound image and the shadow image; inputting, into a first decoder and a second decoder, an output acquired from an encoder in response to inputting the combined image into the encoder; and executing training of the encoder, the first decoder, and the second decoder, based on: reconfigured error between an output image of a coupling function and the combined image, the coupling function being configured to combine a first image output from the first decoder with a second image output from the second decoder, and an error function between an area in the first image and the shadow in the shadow image.

公开(公告)号：US20200042876A1

公开(公告)日：2020-02-06

申请号：US16653236

申请日：2019-10-15

Applicant: FUJITSU LIMITED

Inventor： TAKASHI KATOH ,  Kento UEMURA ,  Suguru YASUTOMI ,  Toshio Endoh ,  Koji MARUHASHI

IPC: G06N3/08 ,  G06K9/62 ,  G05B23/02

Abstract: A non-transitory computer-readable recording medium records an estimation program causing a computer to execute processing which includes: calculating a reconfiguration error from an input result value and a reconfiguration value that is estimated by a first estimator, which estimates a parameter value from a result value learned on a basis of past data, and a second estimator, which estimates a result value from a parameter value, by using a specific result value or a neighborhood result value in a neighborhood of the specific result value; searching for a first result value that minimizes a sum of a substitute error that is calculated from the input result value and the specific result value and the reconfiguration error; and outputting a parameter value that is estimated from the first result value by using the first estimator.