Method and device for training a machine learning routine for controlling a technical system

A device for training a machine learning routine for controlling a technical system, the device comprising one or more processors, application specific integrated circuits (ASIC), digital signal processors (DSP), field programmable gate arrays (FPGA), or any combination thereof and one or more memories, the device configured to:

• input a sequence of first training data comprising first sensor data of a technical system for the machine learning routine;
• train the machine learning routine with the first training data, wherein a plurality of learning parameters of the machine learning routine is set by the training;
• determine a value distribution of the learning parameters during the training;
• generate a continuation signal as a function of the determined value distribution of the learning parameters;
• and
• continue the training, depending on the continuation signal, with a further sequence of the first training data or with other training data requested for the training.

A non-transitory machine-readable storage medium having a computer program that when executed by a processor causes the processor to:

• input a sequence of first training data comprising first sensor data of a technical system for the machine learning routine;
• train the machine learning routine with the first training data, wherein a plurality of learning parameters of the machine learning routine is set by the training;
• determine a value distribution of the learning parameters during the training;
• generate a continuation signal as a function of the determined value distribution of the learning parameters;
• and
• continue the training, depending on the continuation signal, with a further sequence of the first training data or with other training data requested for the training.