# US 2022/0033110 A1 · Mitigating damage to multi-layer networks

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

- Number: US 2022/0033110 A1
- Document ID: US20220033110A1
- Status: Published application
- Kind: A1
- Application number: 17/377,281
- Filed: 2021-07-15
- Published: 2022-02-03
- Earliest claimed: 2020-07-29
- Assignees: The Boeing Company
- Applicants: The Boeing Company
- Inventors: Richard A. Effler; Alexander S. Burch
- Web page: https://patentgrep.com/patent/US20220033110A1

## CPC

- B64G 1/546: {shielding electronic equipment}
- G06F 18/214: Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- G06N 3/0464: Convolutional networks [CNN, ConvNet]
- G06N 3/08: Learning methods
- G06N 3/082: modifying the architecture, e.g. adding, deleting or silencing nodes or connections
- G06N 3/09: Supervised learning
- G06V 10/82: using neural networks
- G06V 20/52: Surveillance or monitoring of activities, e.g. for recognising suspicious objects (recognising microscopic objects G06V20/69)
- B64D 45/00: Aircraft indicators or protectors not otherwise provided for
- B64G: COSMONAUTICS; VEHICLES OR EQUIPMENT THEREFOR
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- B64D: EQUIPMENT FOR FITTING IN OR TO AIRCRAFT; FLIGHT SUITS; PARACHUTES; ARRANGEMENT OR MOUNTING OF POWER PLANTS OR PROPULSION TRANSMISSIONS IN AIRCRAFT

## Abstract

Software-based solutions may mitigate physical damage to multi-layer networks, such as neural networks having shortcut (residual) connections. An example includes: providing a multi-layer network comprising a plurality of nodes; for each of a plurality of training cases: determining a set of dropout nodes, based at least on a damage model having a probability of a node being selected for dropout that is based at least on a target operating environment of the multi-layer network, wherein the probability of a node being selected is spatially correlated; and training the multi-layer network with the determined set of dropout nodes disabled (with a different set of dropout nodes for different training cases). In some examples the damage model involves expected physical radiation damage to a computing device hosting the multi-layer network, such as on board an aircraft or an earth-orbiting satellite. Thus, multiple degrees of expected damage may be addressed.

## Claims

### Claim 1

A method of mitigating damage to multi-layer networks, the method comprising:
providing a multi-layer network comprising a plurality of nodes arranged into a plurality of layers;
receiving a plurality of training cases for training the multi-layer network; and
for each training case of the plurality of training cases:
determining a set of dropout nodes within the multi-layer network, based at least on a damage model having a probability of a node being selected for dropout that is based at least on a target operating environment of the multi-layer network, wherein the probability of a node being selected for dropout is spatially correlated; and
training the multi-layer network with the determined set of dropout nodes disabled, wherein a different set of dropout nodes is determined for different training cases.

### Claim 2 (depends on 1)

The method of claim 1, wherein the multi-layer network comprises a neural network, and wherein the plurality of layers comprises an input layer, an output layer, and at least one hidden layer.

### Claim 3 (depends on 1)

The method of claim 1, wherein the multi-layer network comprises at least one shortcut connection among different ones of the plurality of layers.

### Claim 4 (depends on 1)

The method of claim 1, wherein the damage model comprises damage to a computing device hosting the multi-layer network, the damage of a type selected from a list consisting of:
radiation damage, virus, and cyber-attack.

### Claim 5 (depends on 1)

The method of claim 1, further comprising:
operating the multi-layer network in the target operating environment by:
receiving input data by the multi-layer network; and
based at least receiving input data by the multi-layer network, generating an output by the multi-layer network.

### Claim 6 (depends on 1)

The method of claim 1, wherein the probability approximates an inverse Gaussian distribution.

### Claim 7 (depends on 1)

The method of claim 1, wherein the target operating environment comprises at least one operating environment selected from a list consisting of:
a spacecraft, an aircraft and an earth-orbiting satellite.

### Claim 8

A system for mitigating damage to multi-layer networks, the system comprising:
one or more processors; and
a memory storing instructions that, when executed by the one or more processors, cause the one or more processors to perform operations comprising:
providing a multi-layer network comprising a plurality of nodes arranged into a plurality of layers;
receiving a plurality of training cases for training the multi-layer network; and
for each training case of the plurality of training cases:
determining a set of dropout nodes within the multi-layer network, based at least on a damage model having a probability of a node being selected for dropout that is based at least on a target operating environment of the multi-layer network, wherein the probability of a node being selected for dropout is spatially correlated; and
training the multi-layer network with the determined set of dropout nodes disabled, wherein a different set of dropout nodes is determined for different training cases.

### Claim 9 (depends on 8)

The system of claim 8, wherein the multi-layer network comprises a neural network, and wherein the plurality of layers comprises an input layer, an output layer, and at least one hidden layer.

### Claim 10 (depends on 8)

The system of claim 8, wherein the multi-layer network comprises at least one shortcut connection among different ones of the plurality of layers.

### Claim 11 (depends on 8)

The system of claim 8, wherein the damage model comprises damage to a computing device hosting the multi-layer network, the damage of a type selected from a list consisting of:
radiation damage, virus, and cyber-attack.

### Claim 12 (depends on 8)

The system of claim 8, wherein the operations further comprise:
operating the multi-layer network in the target operating environment by:
receiving input data by the multi-layer network; and
based at least receiving input data by the multi-layer network, generating an output by the multi-layer network.

### Claim 13 (depends on 8)

The system of claim 8, wherein the probability approximates an inverse Gaussian distribution.

### Claim 14 (depends on 8)

The system of claim 8, wherein the target operating environment comprises at least one operating environment selected from a list consisting of:
a spacecraft, an aircraft and an earth-orbiting satellite.

### Claim 15

A computer program product, comprising a computer usable medium having a computer readable program code embodied therein, the computer readable program code adapted to be executed to implement a method for mitigating damage to multi-layer networks, the method comprising:
providing a multi-layer network comprising a plurality of nodes arranged into a plurality of layers;
receiving a plurality of training cases for training the multi-layer network; and
for each training case of the plurality of training cases:
determining a set of dropout nodes within the multi-layer network, based at least on a damage model having a probability of a node being selected for dropout that is based at least on a target operating environment of the multi-layer network, wherein the probability of a node being selected for dropout is spatially correlated; and
training the multi-layer network with the determined set of dropout nodes disabled, wherein a different set of dropout nodes is determined for different training cases.

### Claim 16 (depends on 15)

The computer program product of claim 15, wherein the multi-layer network comprises a neural network, and wherein the plurality of layers comprises an input layer, an output layer, and at least one hidden layer.

### Claim 17 (depends on 15)

The computer program product of claim 15, wherein the multi-layer network comprises at least one shortcut connection among different ones of the plurality of layers.

### Claim 18 (depends on 15)

The computer program product of claim 15, wherein the damage model comprises damage to a computing device hosting the multi-layer network, the damage of a type selected from a list consisting of:
radiation damage, virus, and cyber-attack.

### Claim 19 (depends on 15)

The computer program product of claim 15, wherein the method further comprises:
operating the multi-layer network in the target operating environment by:
receiving input data by the multi-layer network; and
based at least receiving input data by the multi-layer network, generating an output by the multi-layer network.

### Claim 20 (depends on 15)

The computer program product of claim 15, wherein the probability approximates an inverse Gaussian distribution.
