US 11,810,338 B2
Machine learning model for image recognition used in autonomous vehicles
Shintaro Fukushima, Tokyo (JP); and Takeyuki Sasai, Tokyo (JP)
Assigned to TOYOTA JIDOSHA KABUSHIKI KAISHA, Toyota (JP)
Filed by TOYOTA JIDOSHA KABUSHIKI KAISHA, Toyota (JP)
Filed on Apr. 16, 2021, as Appl. No. 17/232,405.
Claims priority of application No. 2020-076792 (JP), filed on Apr. 23, 2020.
Prior Publication US 2021/0331693 A1, Oct. 28, 2021
Int. Cl. G06N 20/00 (2019.01); G06T 7/00 (2017.01); G06V 10/764 (2022.01); G06V 10/774 (2022.01); B60W 60/00 (2020.01); G06V 20/56 (2022.01); G06F 18/214 (2023.01); G06V 10/82 (2022.01)
CPC G06V 10/764 (2022.01) [B60W 60/001 (2020.02); G06F 18/214 (2023.01); G06V 10/774 (2022.01); G06V 10/7747 (2022.01); G06V 10/82 (2022.01); G06V 20/56 (2022.01); B60W 2420/403 (2013.01)] 11 Claims
OG exemplary drawing
 
1. An information processing system comprising an information processing device and a control device, the information processing device including:
a first processor that is configured to make a determination as to whether each of a plurality of change amounts corresponding to each of a plurality of second output data relative to each of a plurality of first output data is not more than a pre-specified threshold,
the plurality of first output data being obtained from a trained model by input of a plurality of first input data to the trained model, the plurality of first input data being represented as a first set of points in a space with the same number of dimensions as the first input data, and the trained model being trained beforehand,
the plurality of second output data being obtained from the trained model by input of a plurality of second input data to the trained model, the plurality of second input data being represented as a second set of points in the space with the same number of dimensions as the second input data, and each second input data is obtained by applying a perturbation of a specified perturbation amount, for which the corresponding change amount is determined to be not more than the threshold, to the corresponding first input data such that the each second data is contained in a hypersphere whose radius is the specified perturbation amount and which is centered at a point representing the corresponding first input data;
obtaining a union of a plurality of hyperspheres each of which corresponds to the hypersphere that contains the each second data;
output the union of the plurality of hyperspheres as a range information,
and the control device including:
a second processor that is configured to acquire third input data to be inputted to the trained model;
make a determination as to whether the third input data is contained in the union of the plurality of hyperspheres represented by the range information outputted by the information processing device; and
output the third input data when the third input data is determined to be not contained in the union of the plurality of hyperspheres,
wherein
the first processor is configured to receive the third input data, which is determined to be not contained in the union of the plurality of hyperspheres, and make the determination thereof with the third input data being added to the plurality first input data such that the range information is updated in correspondence with the third input data being added to the plurality of the first data, and
the first processor is configured to output the third input data and a perturbation amount for which the corresponding change amount is determined to be not more than the threshold.