Model calculation unit and control unit for calculating a multilayer perceptron model with feedforward and feedback

摘要

The invention relates to a model calculation unit (22) for calculating a multilayer perceptron model, wherein the model calculation unit (22) is designed and hardwired in hardware, comprising: - a calculation kernel (18) which is designed to generate one or more output variables of an output variable vector (102 ) a neuron layer of the multi-layer perceptron model having a number of neurons dependent on one or more inputs of an input vector (101); A memory (12) in which a data storage area (122) is provided, in which each neuron layer a data storage section (D) for storing the input quantities of the input quantity vector (101) and a data storage section (D) for storing the outputs of the output size vector (102) assigned; A DMA unit (17) configured to sequentially instruct the arithmetic core (18) to calculate a respective neuron layer based on inputs of the associated input vector (101) and the respective resulting outputs of the output vector (102) into the associated data memory section (D), wherein the data storage section (D) for the at least one of the neuron layers associated input variable vector (101) at least partially comprise the data storage sections (D) of at least two of the output variable vectors (102) of two different neuron layers.