# US 2021/0081784 A1 · Device and computer implemented method for training an artificial neural network

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2021/0081784 A1
- Document ID: US20210081784A1
- Status: Published application
- Kind: A1
- Application number: 16/983,717
- Filed: 2020-08-03
- Published: 2021-03-18
- Earliest claimed: 2019-09-17
- Applicants: Robert Bosch GmbH
- Inventors: Thomas Andy Keller; Anna Khoreva; Max Welling
- Web page: https://patentgrep.com/patent/US20210081784A1

## CPC

- G06N 3/08: Learning methods
- G06N 3/084: Backpropagation, e.g. using gradient descent
- G06F 17/15: Correlation function computation {including computation of convolution operations (arithmetic circuits for sum of products per se, e.g. multiply-accumulators G06F7/5443; digital filters, e.g. FIR, IIR, adaptive filters H03H17/00)}
- G06N 3/045: Combinations of networks
- G06N 3/0464: Convolutional networks [CNN, ConvNet]
- G06N 3/09: Supervised learning
- G06N 3/0985: Hyperparameter optimisation; Meta-learning; Learning-to-learn
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

Device and method for training an artificial neural network, including providing a neural network layer for an equivariant feature mapping having a plurality of output channels, grouping channels of the output channels into a number of distinct groups, wherein the output channels of each individual distinct group are organized into an individual grid defining a spatial location of each of the output channels of the individual distinct group in the grid for the individual distinct group, providing for each of the output channels of each individual distinct group, a distinct normalization function which is defined depending on the spatial location of the output channel in the grid in that this output channel is organized and depending on tunable hyperparameters for the normalization function, determining an output of the artificial neural network depending on a result of each of the distinct normalization functions, training the hyperparameters of the artificial neural network.

## Claims

### Claim 1

A computer implemented method for training an artificial neural network, comprising the following steps:
providing a neural network layer for an equivariant feature mapping having a plurality of output channels;
grouping channels of the plurality of output channels of the neural network layer into a number of distinct groups, wherein the output channels of each individual distinct group of the number of distinct groups are organized into an individual grid defining a spatial location of each of the output channels of the individual distinct group in the grid for the individual distinct group;
providing, for each of the output channels of each individual distinct group of the number of distinct groups, a distinct normalization function, wherein the distinct normalization function is defined depending on the spatial location of the output channel in the grid in that the output channel is organized, and depending on tunable hyperparameters for the normalization function;
determining an output of the artificial neural network depending on a result of each of the distinct normalization functions; and
training the hyperparameters in a training of the artificial neural network;
wherein the providing of the distinct normalization function, includes providing, for each of the output channels of each individual distinct group of the number of distinct groups, a distinct normalization function, a neighborhood function defining a set of output channels, wherein the normalization function for the output channel is defined depending on output channels of the neural network layer in the set of output channels and disregarding other output channels.

### Claim 2 (depends on 1)

The method according to claim 1, wherein the neural network layer is a convolutional neural network layer.

### Claim 3 (depends on 1)

The method according to claim 1, further comprising the following step:
providing a dimension parameter, wherein the grouping includes grouping the channels of the plurality of output channels of the neural network layer into a number of distinct groups each having a size, wherein the size is defined depending on a total number of output channels of the neural network layer, and organizing for each of the distinct groups its output channels into a grid having a dimension according to the dimension parameter.

### Claim 4 (depends on 1)

The method according to claim 1, further comprising the following steps:
determining a rescaling parameter for each of the output channels; and
rescaling the output for each of the output channels after normalization depending on the rescaling parameter for the output channel.

### Claim 5 (depends on 1)

The method according to claim 1, wherein the set of output channels is defined by a distance parameter defining a distance between output channels of the neural network layer, and wherein the training includes training the distance parameter in the training.

### Claim 6 (depends on 5)

The method according to claim 5, wherein the set of output channels is defined by a spatial distance parameter defining a distance between output channels of the neural network layer in the grid, and the training includes training the spatial distance parameter in the training.

### Claim 7 (depends on 1)

The method according to claim 1, further comprising the following steps:
providing characterizing a digital image;
processing the sensor data depending on the so trained artificial neural network to determine an output signal for classifying the sensor data; and
detecting, or localizing or segmenting objects in the sensor data and/or detecting anomalies in the sensor data.

### Claim 8 (depends on 7)

The method according to claim 7, further comprising:
actuating a physical system depending on the output signal or generating input data for a generative adversarial neural network or a variational autoencoder for data synthesis.

### Claim 9 (depends on 1)

The method according to claim 1, wherein the artificial neural network is adapted for segmenting, or classifying, or detecting pedestriants, and/or road signs and/or vehicles in digital images, and the method further comprises the following steps:
collecting a set of digital images from a database;
applying one or more transformations to each of the digital images including mirroring, or rotating, or smoothing, or contrast reduction, to create a modified set of digital images;
creating a first training set including the collected set of digital images, the modified set of digital images, and a set of digital images unrelated to pedestriants, road signs and/or vehicles;
training the artificial neural network in a first stage using the first training set;
creating a second training set for a second stage of training including the first training set and digital images that are incorrectly detected as images depicting pedestriants, road signs and/or vehicles after the first stage of training; and
training the artificial neural network in a second stage using the second training set.

### Claim 10

A device for processing data, comprising:
an input for data;
an output for an output signal;
at least one processor; and
a memory for an artificial neural network;
wherein the device is configured to train the artificial neural network, the device being configured to:
provide a neural network layer for an equivariant feature mapping having a plurality of output channels;
group channels of the plurality of output channels of the neural network layer into a number of distinct groups, wherein the output channels of each individual distinct group of the number of distinct groups are organized into an individual grid defining a spatial location of each of the output channels of the individual distinct group in the grid for the individual distinct group;
provide, for each of the output channels of each individual distinct group of the number of distinct groups, a distinct normalization function, wherein the distinct normalization function is defined depending on the spatial location of the output channel in the grid in that the output channel is organized, and depending on tunable hyperparameters for the normalization function;
determine an output of the artificial neural network depending on a result of each of the distinct normalization functions; and
train the hyperparameters in a training of the artificial neural network;
wherein the provision of the distinct normalization function, includes providing, for each of the output channels of each individual distinct group of the number of distinct groups, a distinct normalization function, a neighborhood function defining a set of output channels, wherein the normalization function for the output channel is defined depending on output channels of the neural network layer in the set of output channels and disregarding other output channels.

### Claim 11

A non-transitory computer readable storage medium on which is stores computer program for training an artificial neural network, the computer program, when executed by a computer, causing the computer to perform the following steps:
providing a neural network layer for an equivariant feature mapping having a plurality of output channels;
grouping channels of the plurality of output channels of the neural network layer into a number of distinct groups, wherein the output channels of each individual distinct group of the number of distinct groups are organized into an individual grid defining a spatial location of each of the output channels of the individual distinct group in the grid for the individual distinct group;
providing, for each of the output channels of each individual distinct group of the number of distinct groups, a distinct normalization function, wherein the distinct normalization function is defined depending on the spatial location of the output channel in the grid in that the output channel is organized, and depending on tunable hyperparameters for the normalization function;
determining an output of the artificial neural network depending on a result of each of the distinct normalization functions; and
training the hyperparameters in a training of the artificial neural network;
wherein the providing of the distinct normalization function, includes providing, for each of the output channels of each individual distinct group of the number of distinct groups, a distinct normalization function, a neighborhood function defining a set of output channels, wherein the normalization function for the output channel is defined depending on output channels of the neural network layer in the set of output channels and disregarding other output channels.
