Method for operating a deep neural network
The invention relates to a method for operating a deep neural network, wherein the deep neural network is operated with multiple layers between an input layer and an output layer, and wherein, in addition, at least one classic filter is used in the deep neural network between the input and the output layer. The invention also relates to a device for data processing and to a computer-readable storage medium.
1 . A method for operating a deep neural network, comprising:
operating the deep neural network with multiple layers between an input layer and an output layer, wherein at least one classic filter is used in the deep neural network between the input layer and the output layer in parallel to at least one filter channel;
conducting a training phase of the deep neural network, wherein filter parameters of the at least one classic filter are selected on the basis of empirical values and fixed during the training phase of the deep neural network, and wherein filter parameters of the at least one filter channel are randomly initialized;
feeding output data from the at least one classic filter to a backpropagation network of the deep neural network in addition feature maps extracted by the at least one filter channel; and
conducting an inference phase of the deep neural network for the operation of a driver assistance system of a vehicle, wherein the filter parameters of the at least one classic filter remain fixed during the inference phase of the deep neural network.
2 . The method of claim 1 , comprising: conducting a training phase of the deep neural network, wherein at least one part of filter parameters of the at least one classic filter is changed during the training phase of the deep neural network.
3 . The method of claim 2 , wherein at least one part of the filter parameters of the at least one classic filter is adapted with a lower learning rate than the remaining deep neural network.
4 . The method of claim 1 , wherein the at least one classic filter is operated directly after the input layer and/or in the vicinity of the input layer of the deep neural network.
5 . The method of claim 1 , wherein an output of the at least one classic filter is fed to multiple layers of the deep neural network.
6 . The method of claim 1 , wherein the deep neural network is a convolutional neural network.
7 . The method of claim 2 , wherein the at least one classic filter is operated directly after the input layer and/or in the vicinity of the input layer of the deep neural network.
8 . The method of claim 3 , wherein the at least one classic filter is operated directly after the input layer and/or in the vicinity of the input layer of the deep neural network.
9 . The method of claim 2 , wherein an output of the at least one classic filter is fed to multiple layers of the deep neural network.
10 . The method of claim 3 , wherein an output of the at least one classic filter is fed to multiple layers of the deep neural network.
11 . The method of claim 4 , wherein an output of the at least one classic filter is fed to multiple layers of the deep neural network.
12 . The method of claim 2 , wherein the deep neural network is a convolutional neural network.
13 . The method of claim 3 , wherein the deep neural network is a convolutional neural network.
14 . The method of claim 4 , wherein the deep neural network is a convolutional neural network.
15 . A device for operating a deep neural network, the device having a processor configured to:
operate the deep neural network with multiple layers between an input layer and an output layer, wherein at least one classic filter is used in the deep neural network between the input layer and the output layer in parallel to at least one filter channel;
conduct a training phase of the deep neural network, wherein filter parameters of the at least one classic filter are selected on the basis of empirical values and fixed during the training phase of the deep neural network, and wherein filter parameters of the at least one filter channel are randomly initialized;
feed output data from the at least one classic filter to a backpropagation network of the deep neural network in addition feature maps extracted by the at least one filter channel; and
conduct an inference phase of the deep neural network for the operation of a driver assistance system of a vehicle, wherein the filter parameters of the at least one classic filter remain fixed during the inference phase of the deep neural network.
16 . A non-transitory computer-readable storage medium comprising commands which, when run by a computer, prompt the computer to:
operate a deep neural network with multiple layers between an input layer and an output layer, wherein at least one classic filter is used in the deep neural network between the input layer and the output layer in parallel to at least one filter channel;
conduct a training phase of the deep neural network, wherein filter parameters of the at least one classic filter are selected on the basis of empirical values and fixed during the training phase of the deep neural network, and wherein filter parameters of the at least one filter channel are randomly initialized;
feed output data from the at least one classic filter to a backpropagation network of the deep neural network in addition feature maps extracted by the at least one filter channel; and
conduct an inference phase of the deep neural network for the operation of a driver assistance system of a vehicle, wherein the filter parameters of the at least one classic filter remain fixed during the inference phase of the deep neural network.