Method and device for ascertaining a classification and/or a regression result when missing sensor data

1 . A computer-implemented method for ascertaining a classification and/or a regression result based on the plurality of sensor values, the method comprising the following steps:

ascertaining a plurality of hypotheses regarding a missing sensor value using a machine learning system, wherein the missing sensor value is determined to be missing on the basis of a checksum of the sensor values;

ascertaining a plurality of outputs, each output of the plurality of outputs being ascertained based on a plurality of sensor values and a hypothesis and the output characterizing a classification and/or a regression result, wherein a model is configured to perform the classification and/or the regression, and wherein the plurality of outputs is ascertained by supplying the model with each respective hypothesis in place of the missing sensor value;

providing an aggregation of the plurality of outputs as the classification and/or the regression result; and

providing an actuator configured to be controlled based on the classification and/or based on the regression result and/or based on a dispersion value, the actuator being configured to effect an operation of a technical system.

2 . The method as recited in claim 1 , wherein each sensor value characterizes a pixel of an image and/or each sensor value characterizes a voxel in a 3D image and/or each sensor value characterizes a value of an audio signal and/or each sensor value characterizes a measurement of a piezoelectric sensor.

3 . The method as recited in claim 1 , wherein the machine learning system ascertains each hypothesis based on the plurality of sensor values.

4 . The method as recited in claim 3 , wherein the machine learning system includes a conditional normalizing flow, by which the hypothesis is ascertained based on the plurality of sensor values.

5 . The method as recited in claim 1 , wherein the dispersion value characterizing a dispersion of the plurality of outputs.

6 . The method as recited in claim 5 , wherein an actuator is controlled based on the classification and/or based on the regression result and/or based on the dispersion value.

7 . The method as recited in claim 1 , wherein the machine learning system is trained based on a training data set, the training data set including a plurality of training data, each training datum including a plurality of sensor values, and the training comprises the following steps:

selecting a training datum from the training data set;

selecting a sensor value of the training datum;

training the machine learning system in such a way that the machine learning system ascertains the selected sensor value based on the sensor values of the training datum except for the selected sensor value.

8 . The method as recited in claim 7 , wherein the training datum is randomly selected from the plurality of training data and/or the sensor value is randomly selected from the plurality of sensor values of the training datum.

9 . A control device, which is configured to control an actuator, the control device configured to:

ascertain a plurality of hypotheses regarding a missing sensor value using a machine learning system, wherein the missing sensor value is determined to be missing on the basis of a checksum of the sensor values;

ascertain a plurality of outputs, each output of the plurality of outputs being ascertained based on a plurality of sensor values and a hypothesis and the output characterizing a classification and/or a regression result, wherein a model is configured to perform the classification and/or the regression, and wherein the plurality of outputs is ascertained by supplying the model with each respective hypothesis in place of the missing sensor value;

provide an aggregation of the plurality of outputs as the classification and/or the regression result; and

control the actuator based on the classification and/or based on the regression result, the actuator being configured to effect an operation of a technical system.

10 . A non-transitory machine-readable storage medium on which is stored a computer program for ascertaining a classification and/or a regression result based on the plurality of sensor values, the computer program, when executed by a processor, causing the processor to perform the following steps:

ascertaining a plurality of hypotheses regarding a missing sensor value using a machine learning system, wherein the missing sensor value is determined to be missing on the basis of a checksum of the sensor values;

ascertaining a plurality of outputs, each output of the plurality of outputs being ascertained based on a plurality of sensor values and a hypothesis and the output characterizing a classification and/or a regression result, wherein a model is configured to perform the classification and/or the regression, and wherein the plurality of outputs is ascertained by supplying the model with each respective hypothesis in place of the missing sensor value; and

providing an aggregation of the plurality of outputs as the classification and/or the regression result, wherein an actuator is controlled based on the classification and/or based on the regression result and/or based on a dispersion value, the actuator being configured to effect an operation of a technical system.