Estimating reliability of control data

A computer-implemented method of generating a synthetic directional data instance according to a generative model, the synthetic directional data instance including at least one direction vector associated with a physical object, the method comprising the following steps:

• accessing model data representing the generative model, wherein the generative model is trained to generate synthetic directional data instances representative of a training dataset;
• applying the generative model to generate the synthetic directional data instance, and outputting the generated synthetic directional data instance, wherein the generative model includes a coupling layer transforming an input direction vector to an output direction vector, the coupling layer including a circle transformation and one or more of an unconditional rotation and a conditional rotation, wherein:
• the circle transformation transforms a pair of Cartesian coordinates representing a point on a circle to a further point on the circle, wherein the circle transformation is defined based on parameters of the generative model and/or further Cartesian coordinates;
• the unconditional rotation rotates a direction vector according to parameters of the generative model;
• the conditional rotation rotates a first set of Cartesian coordinates based on a second set of Cartesian coordinates.

A system for estimating a reliability of control data for a computer-controlled system interacting with an environment, wherein the control data is inferred from a model input by a machine learnable control model which is trained on a training dataset, the system comprising:

• a data interface configured to access model data representing a generative model, wherein the generative model is trained to generate synthetic model inputs representative of the training dataset of the control model;
• an input interface configured to obtain the model input, wherein the model input includes at least one direction vector which is extracted from sensor data and which is associated with a component of the computer-controlled system or an object in the environment;
• and
• a processor subsystem configured to apply an inverse of the generative model to the model input to determine a likelihood of the model input being generated according to the generative model, and to determine a reliability measure for the control data of the machine learnable model based on the likelihood;
• wherein the generative model includes a coupling layer transforming an input direction vector to an output direction vector, the coupling layer including a circle transformation and one or more of an unconditional rotation and a conditional rotation, wherein:
• the circle transformation transforms a pair of Cartesian coordinates representing a point on a circle to a further point on the circle, wherein the circle transformation is defined based on parameters of the generative model and/or further Cartesian coordinates;
• the unconditional rotation rotates a direction vector according to parameters of the generative model;
• the conditional rotation rotates a first set of Cartesian coordinates based on a second set of Cartesian coordinates.

A system for generating a synthetic directional data instance according to a generative model, the synthetic directional data instance comprising at least one direction vector associated with a physical object, the system comprising:

• a data interface configured to access model data representing the generative model, wherein the generative model is trained to generate synthetic directional data instances representative of a training dataset;
• a processor subsystem configured to apply the generative model to generate the synthetic directional data instance, and output the generated synthetic directional data instance;
• wherein the generative model includes a coupling layer transforming an input direction vector to an output direction vector, the coupling layer including a circle transformation and one or more of an unconditional rotation and a conditional rotation, wherein:
• the circle transformation transforms a pair of Cartesian coordinates representing a point on a circle to a further point on the circle, wherein the circle transformation is defined based on parameters of the generative model and/or further Cartesian coordinates;
• the unconditional rotation rotates a direction vector according to parameters of the generative model;
• the conditional rotation rotates a first set of Cartesian coordinates based on a second set of Cartesian coordinates.

A non-transitory computer-readable medium on which is stored instructions for estimating a reliability of control data for a computer-controlled system interacting with an environment, wherein the control data is inferred from a model input by a machine learnable control model which is trained on a training dataset, the instructions, when executed by a processor system, causing the processor system to perform the following steps:

• accessing model data representing a generative model,
• wherein the generative model is trained to generate synthetic model inputs representative of the training dataset of the control model;
• obtaining the model input, wherein the model input includes at least one direction vector which is extracted from sensor data and which is associated with a component of the computer-controlled system or an object in the environment;
• applying an inverse of the generative model to the model input to determine a likelihood of the model input being generated according to the generative model, and determining a reliability measure for the control data of the machine learnable model based on the likelihood, wherein the generative model includes a coupling layer transforming an input direction vector to an output direction vector, the coupling layer including a circle transformation and one or more of an unconditional rotation and a conditional rotation, wherein:
• the circle transformation transforms a pair of Cartesian coordinates representing a point on a circle to a further point on the circle, wherein the circle transformation is defined based on parameters of the generative model and/or further Cartesian coordinates;
• the unconditional rotation rotates a direction vector according to parameters of the generative model;
• the conditional rotation rotates a first set of Cartesian coordinates based on a second set of Cartesian coordinates.