US 12,333,406 B2
Method and device for classifying sensor data and for ascertaining an activation signal for activating an actuator
Frank Schmidt, Leonberg (DE); and Torsten Sachse, Renningen (DE)
Assigned to ROBERT BOSCH GMBH, Stuttgart (DE)
Appl. No. 17/295,765
Filed by Robert Bosch GmbH, Stuttgart (DE)
PCT Filed Nov. 28, 2019, PCT No. PCT/EP2019/082838
§ 371(c)(1), (2) Date May 20, 2021,
PCT Pub. No. WO2020/126379, PCT Pub. Date Jun. 25, 2020.
Claims priority of application No. 102018222346.7 (DE), filed on Dec. 19, 2018; and application No. 102019215120.5 (DE), filed on Oct. 1, 2019.
Prior Publication US 2022/0012560 A1, Jan. 13, 2022
Int. Cl. G06N 3/04 (2023.01); G06N 3/08 (2023.01)
CPC G06N 3/04 (2013.01) [G06N 3/08 (2013.01)] 18 Claims
OG exemplary drawing
 
1. A method for classifying input signals, which were ascertained as a function of an output signal of a sensor, using a neural network, the method comprising:
providing the neural network, the neural network including a scaling layer;
mapping, by the scaling layer, an input signal (z4) present at an input of the scaling layer onto an output signal (z5) present at an output of the scaling layer in such a way that the mapping corresponds to a projection of the input signal (z4) present at the input of the scaling layer onto a predefinable value range, parameters being predefinable, which characterize the mapping;
wherein the predefinable value range is a ball, which is characterized by a predefinable center (c) of the ball and a predefinable radius (ρ) of the ball;
wherein, in a training phase, the predefinable center (c) of the ball and the predefinable radius (ρ) of the ball are adapted as a function of training the neural network, an adaptation of the predefinable center (c) of the ball and the predefinable radius (ρ) of the ball occurring during the training as a function of an output signal (y) of the neural network when an input signal (x) of the neural network is supplied and as a function of an associated desired output signal, and the adaptation of the predefinable center (c) of the ball and the predefinable radius (ρ) of the ball occurring as a function of an ascertained gradient which is dependent on the output signal (y) of the neural network and the associated desired output signal.