公开(公告)号：US20230029988A1

公开(公告)日：2023-02-02

申请号：US17791742

申请日：2020-01-14

Applicant: NEC Corporation

Inventor： Kosuke YOSHIDA

IPC: G06V10/776 ,  G06V10/774 ,  G06V40/16

Abstract: A learning device performs learning a facial recognition model so as to further reduce a triplet loss that uses a first facial image, a second facial image that is a candidate for an adversarial example of a same person as the first facial image, and a third facial image that is a candidate for an adversarial example of a different person than the first facial image.

公开(公告)号：US20220261642A1

公开(公告)日：2022-08-18

申请号：US17630643

申请日：2019-08-02

Applicant: NEC Corporation

Inventor： Kosuke YOSHIDA

IPC: G06N3/08 ,  G06N7/00

Abstract: An adversarial example detection system capable of detecting adversarial examples at a low computational cost is provided. The preparation unit 100 calculates an inverse matrix of a Gram matrix that is used in a process of approximating a deep learner to a Gaussian process. The output distribution calculation unit 222 calculates mean and variance of output values that are numerical values used for class determination for each class by using the inverse matrix of the Gram matrix, for each input observation data. The probabilistic margin calculation unit 223 calculates a probabilistic margin that is an index of variability of the output values based on the mean and variance of the output values, for each input observation data. The adversarial example detection unit 224 detects the adversarial example from the input observation data based on the probabilistic margin calculated for each input observation data.

公开(公告)号：US20220083039A1

公开(公告)日：2022-03-17

申请号：US17421521

申请日：2019-01-18

Applicant: NEC Corporation

Inventor： Kosuke YOSHIDA

IPC: G05B23/02 ,  G06N3/04 ,  G06K9/62

Abstract: An abnormality detection apparatus (200) includes storage means (210) for storing a learned self-encoder (211) including predetermined number of two or more of elements as input layers, extraction means (220) for extracting a target data group of a predetermined period including data pieces from time series data measured by one or more sensors, the number of the data pieces being the predetermined number, conversion means (230) for converting the target data group into multi-dimensional vector data including the predetermined number of elements, identifying means (240) for identifying a time period in which there may be a cause of an abnormality from the predetermined period based on a difference between output vector data having the predetermined number of elements obtained by inputting the multi-dimensional vector data to the self-encoder (211) and the multi-dimensional vector data, and output means (250) for outputting abnormality detection information including the identified time period.

公开(公告)号：US20210342396A1

公开(公告)日：2021-11-04

申请号：US16980234

申请日：2018-03-14

Applicant: NEC Corporation

Inventor： Jun NISHIOKA ,  Yoshiaki SAKAE ,  Kazuhiko ISOYAMA ,  Etsuko ICHIHARA ,  Kosuke YOSHIDA

IPC: G06F16/901 ,  G06F16/33 ,  G06F16/31 ,  G06F16/28 ,  G06F3/0484

Abstract: To enable a user to easily recognize temporal order of elements included in a retrieval sentence, a retrieval sentence utilization device 10 includes: a retrieval sentence division unit 11 for dividing a retrieval sentence into a plurality of retrieval contents each of which includes an event; and a directed graph generation unit 12 for generating, from each of the retrieval contents, a subtree in which the event is an edge and a source of the event and an object of the event are nodes, and integrating a plurality of subtrees generated from the retrieval contents to generate a directed graph, wherein the directed graph generation unit 12 places the plurality of subtrees in the directed graph according to occurrence order of events corresponding to the plurality of subtrees.

公开(公告)号：US20210042412A1

公开(公告)日：2021-02-11

申请号：US16976311

申请日：2018-03-01

Applicant: NEC CORPORATION

Inventor： Kazuhiko ISOYAMA ,  Yoshiaki SAKAE ,  Jun NISHIOKA ,  Etsuko ICHIHARA ,  Kosuke YOSHIDA

IPC: G06F21/56 ,  G06N20/20

Abstract: An information processing apparatus (2000) classifies each event that occurred in a target apparatus to be determined (10) either as an event (event of a first class) that also occurs in a standard apparatus (20) or as an event (event of a second class) that does not occur in the standard apparatus (20). Herein, a first model used for a determination with respect to an event that also occurs in the standard apparatus (20) and a second model used for a determination with respect to an event that does not occur in the standard apparatus (20) are used as models for determining whether an event that occurs in a target apparatus to be determined (10) is a target for warning. The information processing apparatus (2000) performs learning of the first model using an event of the first class. Further, the information processing apparatus (2000) performs learning of the second model using an event of the second class.