# US 2021/0241113 A1 · Methods and systems for reducing dimensionality in a reduction and prediction framework

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

- Number: US 2021/0241113 A1
- Document ID: US20210241113A1
- Status: Published application
- Kind: A1
- Application number: 17/169,359
- Filed: 2021-02-05
- Published: 2021-08-05
- Earliest claimed: 2020-02-05
- Assignees: Eigen Innovations Inc.
- Applicants: Eigen Innovations Inc.
- Inventors: Ryan FINN
- Web page: https://patentgrep.com/patent/US20210241113A1

## CPC

- G06T 7/0004: {Industrial image inspection}
- G06F 18/217: {Validation; Performance evaluation; Active pattern learning techniques}
- G06F 18/23213: with fixed number of clusters, e.g. K-means clustering
- G06F 18/2413: based on distances to training or reference patterns
- G06N 3/045: Combinations of networks
- G06N 3/0499: Feedforward networks
- G06N 3/082: modifying the architecture, e.g. adding, deleting or silencing nodes or connections
- G06N 3/09: Supervised learning
- G06N 5/022: Knowledge engineering; Knowledge acquisition
- G06N 20/00: Machine learning
- G06V 10/763: {Non-hierarchical techniques, e.g. based on statistics of modelling distributions}
- G06V 10/764: using classification, e.g. of video objects
- G06V 10/7715: {Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods}
- G06V 10/82: using neural networks
- G06N 3/042: Knowledge-based neural networks; Logical representations of neural networks
- G06T 2207/20081: Training; Learning
- G06T 2207/20084: Artificial neural networks [ANN]
- G06T 2207/30108: Industrial image inspection
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- 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

## Abstract

Method and system that includes receiving an sensed data point from an industrial process; applying a mapping model to map the sensed data point to a respective embedding that has reduced dimensionality relative to the sensed data point; determining, based on a comparison of the respective embedding to prior embeddings, if the mapping model needs to be updated or not. When the mapping model needs to be updated, applying manifold learning to learn an updated set of model parameters for the mapping model. When the mapping model does not need to be updated, applying a classification model to the respective embedding to predict a classification for the sensed data point.

## Claims

### Claim 1

A computer implemented method comprising:
receiving an sensed data point from an industrial process;
applying a mapping model to map the sensed data point to a respective embedding that has reduced dimensionality relative to the sensed data point;
determining, based on a comparison of the respective embedding to prior embeddings, if the mapping model needs to be updated or not:
when the mapping model needs to be updated, applying manifold learning to learn an updated set of model parameters for the mapping model; and
when the mapping model does not need to be updated, applying a classification model to the respective embedding to predict a classification for the sensed data point.

### Claim 2 (depends on 1)

The computer implemented method of claim 1 wherein the method further comprises:
when the mapping model needs to be updated: updating the classification model based on embeddings generated based on the updated set of model parameters.

### Claim 3 (depends on 2)

The computer implemented model of claim 2 wherein the method further comprises:
when the mapping model needs to be updated: following updating of the mapping model and the classification model, applying the updated mapping model to map the sensed data point to a respective updated embedding, and applying the updated classification model to the respective updated embedding to predict a classification for the sensed data point.

### Claim 4 (depends on 1)

The method of claim 1 wherein applying manifold learning to learn the updated set of model parameters is performed off-line relative to the industrial process and for the mapping model; and applying the classification model to predict the classification for the sensed data point is performed on-line relative to the industrial process.

### Claim 5 (depends on 1)

The method of claim 1 wherein the sensed data point comprises image data.

### Claim 6 (depends on 5)

The method of claim 5 wherein the classification model is configured to predict a classification as to the quality of an output product of the industrial process based on image data that captures a representation of the output product.

### Claim 7 (depends on 1)

The method of claim 1 wherein determining if the mapping model needs to be updated or not comprises determining if the respective embedding falls within one of k-clusters generated using a k-means clustering algorithm in respect of a training dataset.

### Claim 8

A data processing system comprising one or more processors and one or more non-transitory storage mediums storing software instructions, that, when executed by the one or more processors cause the data processing system to perform a method comprising:
receiving an sensed data point from an industrial process;
applying a mapping model to map the sensed data point to a respective embedding that has reduced dimensionality relative to the sensed data point;
determining, based on a comparison of the respective embedding to prior embeddings, if the mapping model needs to be updated or not:
when the mapping model needs to be updated, applying manifold learning to learn an updated set of model parameters for the mapping model; and
when the mapping model does not need to be updated, applying a classification model to the respective embedding to predict a classification for the sensed data point.

### Claim 9 (depends on 8)

The data processing system of claim 8 wherein the method further comprises:
when the mapping model needs to be updated: updating the classification model based on embeddings generated based on the updated set of model parameters.

### Claim 10 (depends on 9)

The data processing system of claim 9 wherein the method further comprises:
when the mapping model needs to be updated: following updating of the mapping model and the classification model, applying the updated mapping model to map the sensed data point to a respective updated embedding, and applying the updated classification model to the respective updated embedding to predict a classification for the sensed data point.

### Claim 11 (depends on 8)

The data processing system of claim 8 wherein applying manifold learning to learn the updated set of model parameters is performed off-line relative to the industrial process and for the mapping model; and applying the classification model to predict the classification for the sensed data point is performed on-line relative to the industrial process.

### Claim 12 (depends on 8)

The data processing system of claim 8 wherein the sensed data point comprises image data.

### Claim 13 (depends on 12)

The data processing system of claim 12 wherein the classification model is configured to predict a classification as to the quality of an output product of the industrial process based on image data that captures a representation of the output product.

### Claim 14 (depends on 8)

The data processing system of claim 8 wherein determining if the mapping model needs to be updated or not comprises determining if the respective embedding falls within one of the k-clusters generated using a k-means clustering algorithm in respect of a training dataset.

### Claim 15

A method comprising:
determining if a mapping model is accurate for mapping out-of-sample (OoS) data points to respective embeddings;
if the mapping model is determined to be accurate, using the mapping model to map the OoS data points to the respective embeddings and a second model to map the embeddings to a prediction; and
if the mapping model is determined not to be accurate:
determining new mapping model parameters for the mapping model using manifold learning;
determining new second model parameters for the second model;
using the mapping model with the updated mapping model parameters to map the OoS data points to new respective embeddings and using the second model with the updated second model parameters to map the new respective embeddings to a prediction.

### Claim 16 (depends on 15)

The method of claim 15 wherein the second model generates an image classification.

### Claim 17 (depends on 16)

The method of claim 16 wherein the second model generates a parameter prediction for an industrial process.