Method, device and computer program for creating a deep neural network

A non-transitory machine-readable memory element on which is stored a computer program for creating a deep neural network, the deep neural network including a plurality of layers and connections having weights, and the weights in the created deep neural network being able to assume only predefinable discrete values from a predefinable list of discrete values, the computer program, when executed by a computer, causing the computer to perform the following steps:

• providing at least one training input variable for the deep neural network;
• ascertaining a variable characterizing a cost function, the variable characterizing the cost function including a first variable, which characterizes a deviation of an output variable of the deep neural network ascertained as a function of the provided training input variable relative to a predefinable setpoint output variable, and the variable characterizing the cost function further including at least one penalization variable which characterizes a deviation of a value of one of the weights from at least one of at least two of the predefinable discrete values;
• training the deep neural network to detect an object as a function of the training input variable of the deep neural network, at least one value of one of the weights being adapted during the training of the deep neural network as a function of the variable characterizing the cost function;
• and
• mapping values of the weights on respective discrete values contained in the predefinable list.

A device configured to create a deep neural network, the deep neural network including a plurality of layers and connections having weights, and the weights in the created deep neural network being able to assume only predefinable discrete values from a predefinable list of discrete values, the device comprising a processor, wherein the processor is programmed, by which programming of the processor, the processor is configured to:

• provide at least one training input variable for the deep neural network;
• ascertain a variable characterizing a cost function, the variable characterizing the cost function including a first variable, which characterizes a deviation of an output variable of the deep neural network ascertained as a function of the provided training input variable relative to a predefinable setpoint output variable, and the variable characterizing the cost function further including at least one penalization variable which characterizes a deviation of a value of one of the weights from at least one of at least two of the predefinable discrete values;
• train the deep neural network to detect an object as a function of the training input variable of the deep neural network, at least one value of one of the weights being adapted during the training of the deep neural network as a function of the variable characterizing the cost function;
• and
• map values of the weights on respective discrete values contained in the predefinable list.