# US 2026/0228571 A1 · Noise-resistant neural network inferencing

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

- Number: US 2026/0228571 A1
- Document ID: US20260228571A1
- Status: Published application
- Kind: A1
- Application number: 19/060,484
- Filed: 2025-02-21
- Published: 2026-08-06
- Earliest claimed: 2025-02-05
- Assignees: NVIDIA Corporation
- Applicants: NVIDIA Corporation
- Inventors: Chong Yu
- Web page: https://patentgrep.com/patent/US20260228571A1

## CPC

- G06N 3/063: using electronic means
- G06N 5/04: Inference or reasoning models
- G06N 3/044: Recurrent networks, e.g. Hopfield networks
- G06N 3/045: Combinations of networks
- G06N 3/047: Probabilistic or stochastic networks
- G06N 3/08: Learning methods
- G06N 3/084: Backpropagation, e.g. using gradient descent
- G06N 3/088: Non-supervised learning, e.g. competitive learning
- G06N 20/00: Machine learning
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

Apparatuses, systems, and techniques to perform inferencing using one or more neural networks with improved noise immunity are disclosed. In at least one embodiment, one or more neural networks may generate inferences, where these inferences may include measurably consistent information from the same input data irrespective of changes in environmental conditions.

## Claims

### Claim 1

A processor comprising:
one or more circuits to use one or more neural networks to generate measurably consistent information from multiple inferences using same information to be inferenced irrespective of changes in environmental conditions.

### Claim 2 (depends on 1)

The processor of claim 1, wherein the multiple inferences comprise a baseline inference and one or more additional inferences, wherein the baseline inference uses the information unperturbed by environmental conditions, and wherein respective inferences of the one or more additional inferences use the information perturbed by environmental conditions.

### Claim 3 (depends on 1)

The processor of claim 1, wherein the measurably consistent information comprises information of respective inference results of the multiple inferences that differ within a variance threshold.

### Claim 4 (depends on 1)

The processor of claim 1, wherein to generate the measurably consistent information the one or more circuits use the one or more neural networks to postprocess respective inferences generated using the information according to a dynamic compensation model and one or more metrics of an execution environment.

### Claim 5 (depends on 1)

The processor of claim 1, wherein the one or more neural networks implement a baseline neural network model and a noise-resistant neural network model, and wherein the noise-resistant neural network model is adversarially trained using inferences of the baseline neural network model.

### Claim 6 (depends on 5)

The processor of claim 5, wherein the noise-resistant neural network model is adversarially trained according to data set perturbed using a noise simulation model modulated by temporal variance.

### Claim 7 (depends on 5)

The processor of claim 5, wherein the adversarially training is performed according to respective differences between inferences of the noise-resistant neural network model and the inferences of the baseline neural network model, and wherein the respective differences are processed by a noise defense model implemented by the one or more neural networks.

### Claim 8

A system, comprising:
one or more processors to cause one or more neural networks to generate measurably consistent information from multiple inferences using same information to be inferenced irrespective of changes in environmental conditions.

### Claim 9 (depends on 8)

The system of claim 8, wherein the multiple inferences comprise a baseline inference and one or more additional inferences, wherein the baseline inference uses the information unperturbed by environmental conditions, and wherein respective inferences of the one or more additional inferences use the information perturbed by environmental conditions.

### Claim 10 (depends on 8)

The system of claim 8, wherein the measurably consistent information comprises information of respective inference results of the multiple inferences that differ within a variance threshold.

### Claim 11 (depends on 8)

The system of claim 8, wherein to generate the measurably consistent information the one or more processors cause the one or more neural networks to postprocess respective inferences generated using the information according to a dynamic compensation model and one or more metrics of an execution environment.

### Claim 12 (depends on 8)

The system of claim 8, wherein the one or more neural networks implement a baseline neural network model and a noise-resistant neural network model, and wherein the noise-resistant neural network model is adversarially trained using inferences of the baseline neural network model.

### Claim 13 (depends on 12)

The system of claim 12, wherein the noise-resistant neural network model is adversarially trained according to data set perturbed using a noise simulation model modulated by temporal variance.

### Claim 14 (depends on 12)

The system of claim 12, wherein the adversarially training is performed according to respective differences between inferences of the noise-resistant neural network model and the inferences of the baseline neural network model, and wherein the respective differences are processed by a noise defense model implemented by the one or more neural networks.

### Claim 15

A method, comprising:
causing, by circuitry of one or more processors, one or more neural networks to generate measurably consistent information from multiple inferences using same information to be inferenced irrespective of changes in environmental conditions.

### Claim 16 (depends on 15)

The method of claim 15, wherein the multiple inferences comprise a baseline inference and one or more additional inferences, wherein the baseline inference uses the information unperturbed by environmental conditions, and wherein respective inferences of the one or more additional inferences use the information perturbed by environmental conditions.

### Claim 17 (depends on 15)

The method of claim 15, wherein the measurably consistent information comprises information of respective inference results of the multiple inferences that differ within a variance threshold.

### Claim 18 (depends on 15)

The method of claim 15, wherein generating the measurably consistent information comprises postprocessing respective inferences generated using the information according to a dynamic compensation model and one or more metrics of an execution environment.

### Claim 19 (depends on 15)

The method of claim 15, wherein the one or more neural networks implement a baseline neural network model and a noise-resistant neural network model, and wherein the noise-resistant neural network model is adversarially trained using inferences of the baseline neural network model.

### Claim 20 (depends on 19)

The method of claim 19, wherein the noise-resistant neural network model is adversarially trained according to data set perturbed using a noise simulation model modulated by temporal variance, and wherein the adversarially training is performed according to respective differences between inferences of the noise-resistant neural network model and the inferences of the baseline neural network model, and wherein the respective differences are processed by a noise defense model implemented by the one or more neural networks.
