Training method for a generator neural network imposing data equivariances

A method to generate further training data for a machine learnable model, the method comprising the following steps:

• obtaining an initial training set for the machine learnable model, the initial training set including measured sensor data obtained from a sensor;
• training a generator network from the initial training set by:
• generating synthesized sensor data using the generator neural network,
• optimizing the discriminator network to distinguish between the measured sensor data and the synthesized sensor data,
• optimizing the generator network to generate the synthesized sensor data which is indistinguishable from the measured sensor data by the discriminator network;
• wherein:
• (i) a fidelity destroying transformation is defined configured to transform measured sensor data to obtain fidelity-destroyed transformed measured sensor data, and the training further includes optimizing the discriminator network to distinguish between the measured sensor data and the fidelity-destroyed transformed measured sensor data, and/or
• (ii) a fidelity preserving transformation is defined configured to transform the measured sensor data to obtain fidelity-preserved transformed measured sensor data, and the training further includes optimizing the discriminator network not to distinguish between the measured sensor data and the fidelity-preserving transformed measured sensor data;
• applying the trained generator network to generate the further training data.

A method to train a machine learnable model, the method comprises the following steps:

• obtaining an initial training set for the machine learnable model;
• generating further training data for the machine learnable model by:
• training a generator network from the initial training set by:
• generating synthesized sensor data using the generator neural network,
• optimizing the discriminator network to distinguish between the measured sensor data and the synthesized sensor data,
• optimizing the generator network to generate the synthesized sensor data which is indistinguishable from the measured sensor data by the discriminator network;
• wherein:
• (i) a fidelity destroying transformation is defined configured to transform measured sensor data to obtain fidelity-destroyed transformed measured sensor data, and the training further includes optimizing the discriminator network to distinguish between the measured sensor data and the fidelity-destroyed transformed measured sensor data, and/or
• (ii) a fidelity preserving transformation is defined configured to transform the measured sensor data to obtain fidelity-preserved transformed measured sensor data, and the training further includes optimizing the discriminator network not to distinguish between the measured sensor data and the fidelity-preserving transformed measured sensor data;
• applying the trained generator network to generate the further training data;
• and
• training and/or testing the machine learnable model at least on the further training data.

A training system for training a generator neural network configured to generate synthesized sensor data, the system comprising:

• a communication interface configured to access a training set of measured sensor data obtained from a sensor;
• and
• a processor system configured to train the generator network together with a discriminator neural network, wherein the training includes:
• generating synthesized sensor data using the generator neural network,
• optimizing the discriminator network to distinguish between the measured sensor data and the synthesized sensor data, and
• optimizing the generator network to generate the synthesized sensor data which is indistinguishable from the measured sensor data by the discriminator network;
• wherein:
• a fidelity destroying transformation is defined configured to transform the measured sensor data to obtain fidelity-destroyed transformed measured sensor data, the training further including optimizing the discriminator network to distinguish between the measured sensor data and the fidelity-destroyed transformed measured sensor data, and/or
• a fidelity preserving transformation is defined configured to transform measured sensor data to obtain fidelity-preserved transformed measured sensor data, the training further including optimizing the discriminator network not to distinguish between the measured sensor data and the fidelity-preserving transformed measured sensor data.

A generator system for a generator neural network arranged to generate synthesized sensor data, the system comprising:

• a processor system arranged to apply a trained generator network, wherein the generator network has been trained with a training system for training the generator neural network to generate synthesized sensor data, the training system including:
• a first communication interface configured to access a training set of measured sensor data obtained from a sensor;
• and
• a first processor system configured to train the generator network together with a discriminator neural network, wherein the training includes:
• generating synthesized sensor data using the generator neural network,
• optimizing the discriminator network to distinguish between the measured sensor data and the synthesized sensor data, and
• optimizing the generator network to generate the synthesized sensor data which is indistinguishable from the measured sensor data by the discriminator network;
• wherein:
• a fidelity destroying transformation is defined configured to transform the measured sensor data to obtain fidelity-destroyed transformed measured sensor data, the training further including optimizing the discriminator network to distinguish between the measured sensor data and the fidelity-destroyed transformed measured sensor data, and/or
• a fidelity preserving transformation is defined configured to transform measured sensor data to obtain fidelity-preserved transformed measured sensor data, the training further including optimizing the discriminator network not to distinguish between the measured sensor data and the fidelity-preserving transformed measured sensor data;
• and
• a communication interface for transmitting or storing the synthesized sensor data.

An autonomous apparatus, comprising:

• a sensor configured to sense an environment of the apparatus and to generate measured sensor data;
• a classifier including a machine learnable model, the classifier being trained to classify an object in the environment of the apparatus from the measured sensor data;
• a controller configured to generate a control signal to control the autonomous apparatus, the controller being configured to generate the control signal at least from the object classified by the classifier;
• an actuator configured to move under the control of the control signal;
• wherein the machine learnable model is trained by:
• obtaining an initial training set for the machine learnable model;
• generating further training data for the machine learnable model by:
• training a generator network from the initial training set by:
• generating synthesized sensor data using the generator neural network,
• optimizing the discriminator network to distinguish between the measured sensor data and the synthesized sensor data,
• optimizing the generator network to generate the synthesized sensor data which is indistinguishable from the measured sensor data by the discriminator network;
• wherein:
• (i) a fidelity destroying transformation is defined configured to transform measured sensor data to obtain fidelity-destroyed transformed measured sensor data, and the training further includes optimizing the discriminator network to distinguish between the measured sensor data and the fidelity-destroyed transformed measured sensor data, and/or
• (ii) a fidelity preserving transformation is defined configured to transform the measured sensor data to obtain fidelity-preserved transformed measured sensor data, and the training further includes optimizing the discriminator network not to distinguish between the measured sensor data and the fidelity-preserving transformed measured sensor data;
• applying the trained generator network to generate the further training data;
• and
• training and/or testing the machine learnable model at least on the further training data.

A non-transitory computer readable medium on which is stored data representing instructions for training a generator neural network configured to generate synthesized sensor data, the instructions, when executed by a processor system, causing the processor system to perform the following steps:

• accessing a training set of measured sensor data obtained from a sensor;
• and
• training the generator network together with a discriminator neural network, the training including:
• generating synthesized sensor data using the generator neural network,
• optimizing the discriminator network to distinguish between the measured sensor data and the synthesized sensor data,
• optimizing the generator network to generate the synthesized sensor data which is indistinguishable from the measured sensor data by the discriminator network;
• wherein:
• (i) a fidelity destroying transformation is defined configured to transform measured sensor data to obtain fidelity-destroyed transformed measured sensor data, and the training further includes optimizing the discriminator network to distinguish between the measured sensor data and the fidelity-destroyed transformed measured sensor data, and/or
• (ii) a fidelity preserving transformation is defined configured to transform the measured sensor data to obtain fidelity-preserved transformed measured sensor data, and the training further includes optimizing the discriminator network not to distinguish between the measured sensor data and the fidelity-preserving transformed measured sensor data.