How to Train AI Models with Fake Data Using Generative Networks
This patent describes a method for training artificial intelligence models using specially generated fake, or 'synthetic,' data created by a generative adversarial network, ensuring the synthetic data is high-quality and safe for training.
Original patent title: “Data model generation using generative adversarial networks”
This patent describes a method for training artificial intelligence models using specially generated fake, or 'synthetic,' data created by a generative adversarial network, ensuring the synthetic data is high-quality and safe for training. Owned by Capital One Services with 25 claims, and it is expected to expire in 2043.
Coverage
What does this patent actually cover?
The patent outlines a system for generating data models, like those used in AI, by first creating synthetic data. A "model optimizer" receives a request to build a data model (ClaimclaimA numbered sentence at the end of a patent that legally defines what the inventor owns. The most important section.Read more → 21). It then sets up computing resources and uses a "generative network" to create a synthetic dataset. This generative network includes a "decoder network" that takes simplified "decoder input data" from a "code space" and transforms it into more complex "decoder output data" in a "sample space" (Claim 21). The generative network is trained to ensure the synthetic data's structure, or "schema," matches that of real "reference data" (Claim 22). Before training, the model optimizer can check the synthetic data's quality by calculating scores for statistical correlation, data similarity, or overall data quality compared to the real data (Claim 23). If the synthetic data meets certain quality standards (Claim 24), the computing resources then use it to train the actual data model. Finally, this trained data model can be used to process real "production data" (Claim 21). For example, a bank could use this to generate fake customer transaction data that looks real but contains no actual customer information, then train a fraud detection AI on this fake data.
The gap
What does this patent NOT cover?
- Does not cover generating synthetic data without using a generative network that specifically includes a decoder network transforming data from a code space to a sample space. (ClaimclaimA numbered sentence at the end of a patent that legally defines what the inventor owns. The most important section.Read more → 21)
- Does not cover training data models directly with real-world, non-synthetic datasets. (ClaimclaimA numbered sentence at the end of a patent that legally defines what the inventor owns. The most important section.Read more → 21)
- Does not cover synthetic data generation where the output data's schema does not match a reference dataset's schema. (ClaimclaimA numbered sentence at the end of a patent that legally defines what the inventor owns. The most important section.Read more → 22)
- Does not cover methods that do not evaluate the synthetic dataset using at least one of a statistical correlation score, a data similarity score, or a data quality score. (ClaimclaimA numbered sentence at the end of a patent that legally defines what the inventor owns. The most important section.Read more → 23)
- Does not cover scenarios where the 'code space' for the decoder input data has a dimensionality equal to or greater than the 'sample space' of the decoder output data. (ClaimclaimA numbered sentence at the end of a patent that legally defines what the inventor owns. The most important section.Read more → 25)
- Does not cover systems that do not employ a 'model optimizer' to manage the request, resource provisioning, and evaluation steps. (ClaimclaimA numbered sentence at the end of a patent that legally defines what the inventor owns. The most important section.Read more → 21)
These exclusions are unique to PatentBrief — derived from the actual claim language, not patent-office boilerplate.
Key facts
What made this novel
The noveltynoveltyThe requirement that an invention be different from anything publicly known before its priority date.Read more → lies in the structured process of generating synthetic data using a specific generative network architecture (decoder/encoder, code/sample space) and then rigorously evaluating its quality against real data's statistical properties and schema before using it for model training. This ensures the synthetic data is both realistic enough for effective training and sufficiently distinct for privacy.
The Patent Drawing
Schematic visualization of the patent's claim structure. Hand-drawn diagrams in progress for each landmark patent.
Where you've seen this
Real-world examples
AI model training platforms for financial services
Healthcare data anonymization tools
Synthetic data generators for fraud detection systems
Machine learning development environments requiring privacy-preserving data
Cloud-based AI/ML training services
Why it matters
The bigger picture
This patent addresses a critical need in AI development: training powerful models without compromising sensitive real-world data. By generating high-quality synthetic data, organizations like Capital One can develop and test AI solutions more rapidly and securely. This approach helps comply with privacy regulations and reduces the risks associated with handling confidential information, enabling innovation in data-sensitive industries.
Filed
May 22, 2023
Market context
Who's building on this
Companies in this space
Capital One Services, the assigneeassigneeThe entity that owns the patent — usually the inventor's employer or a company.Read more →, is actively developing and applying AI and machine learning solutions, particularly in financial services. Other major financial institutions like JPMorgan Chase and Wells Fargo, along with cloud providers such as Amazon Web Services, Google Cloud, and Microsoft Azure, are also investing heavily in synthetic data generation capabilities to train their AI models while adhering to strict data privacy regulations.
Market impact
This technology enables the development of AI models in highly regulated industries by providing a pathway to train on data that mimics real-world characteristics without exposing sensitive information. It helps reduce the time and cost associated with data acquisition and anonymization, fostering innovation in areas like fraud detection, risk assessment, and personalized financial services. This approach could become a standard practice for AI model development where data privacy is paramount.
Claim 1 — Plain English
What this patent covers
The patent outlines a system for generating data models, like those used in AI, by first creating synthetic data. A "model optimizer" receives a request to build a data model (Claim 21). It then sets up computing resources and uses a "generative network" to create a synthetic dataset. This generative network includes a "decoder network" that takes simplified "decoder input data" from a "code space" and transforms it into more complex "decoder output data" in a "sample space" (Claim 21). The generative network is trained to ensure the synthetic data's structure, or "schema," matches that of real "reference data" (Claim 22). Before training, the model optimizer can check the synthetic data's quality by calculating scores for statistical correlation, data similarity, or overall data quality compared to the real data (Claim 23). If the synthetic data meets certain quality standards (Claim 24), the computing resources then use it to train the actual data model. Finally, this trained data model can be used to process real "production data" (Claim 21). For example, a bank could use this to generate fake customer transaction data that looks real but contains no actual customer information, then train a fraud detection AI on this fake data.
The clever bit
The novelty lies in the structured process of generating synthetic data using a specific generative network architecture (decoder/encoder, code/sample space) and then rigorously evaluating its quality against real data's statistical properties and schema before using it for model training. This ensures the synthetic data is both realistic enough for effective training and sufficiently distinct for privacy.
What it does not cover
- Does not cover generating synthetic data without using a generative network that specifically includes a decoder network transforming data from a code space to a sample space. (Claim 21)
- Does not cover training data models directly with real-world, non-synthetic datasets. (Claim 21)
- Does not cover synthetic data generation where the output data's schema does not match a reference dataset's schema. (Claim 22)
- Does not cover methods that do not evaluate the synthetic dataset using at least one of a statistical correlation score, a data similarity score, or a data quality score. (Claim 23)
- Does not cover scenarios where the 'code space' for the decoder input data has a dimensionality equal to or greater than the 'sample space' of the decoder output data. (Claim 25)
- Does not cover systems that do not employ a 'model optimizer' to manage the request, resource provisioning, and evaluation steps. (Claim 21)
Patent timeline
Application submitted to the patent office
Patent enters public domain
PatentBrief Score
Impact Score
Limited data
Citation count
0/40
No citations yet
Claim breadth
17/20
Very broad protection
Recency
0/20
Older than 20 years
Assignee scale
0/20
Independent or smaller assigneeassigneeThe entity that owns the patent — usually the inventor's employer or a company.Read more →
PatentBrief Impact Score — based on citation count, claim breadth, recency, and assignee scale. Not a legal assessment.
Heuristic Value Estimate
What this patent might be worth
$31K – $100K
Midpoint $62K · 16.9 yr remaining · industry ×1.6
Heuristic only — blends forward/backward citation counts, claim scope, time remaining, litigation history, and CPC-derived industry baseline. Real valuations need a professional appraisal.
The original legal language
Original claims
25 claims as filed with the patent office.
Concepts involved
Cite this patent
Walters, A., Key, K., Watson, M., Goodsitt, J., Pham, V., Truong, A., Taylor, K., Farivar, R., & Abad, F. A. T. How to Train AI Models with Fake Data Using Generative Networks (U.S. Patent No. 20,230,297,446). U.S. Patent and Trademark Office. https://patentbrief.org/patent/us/20230297446/data-model-generation-using-generative-adversarial-networks
Auto-generated from the patent record. Double-check author order and the issue date against the official USPTO document before submitting.
Embed
Add this patent to your site
Drop this plain-English patent card into any blog post or article — free, no signup. It always links back to the full breakdown here.
<div data-patentlens-widget data-patent-number="US20230297446"></div> <script src="https://patentbrief.org/embed.js" async></script>
Stay in the loop
Get a weekly digest of new patents.
One email per week. No spam. Unsubscribe anytime.
Keep exploring
Related patents you should know
US 4683195 · 1987
How to Make Billions of Copies of a DNA Segment
This patent describes the Polymerase Chain Reaction (PCR), a method to rapidly create many copies of a specific piece of DNA or RNA, enabling its detection and analysis.
Cetus Corp
US 8697359 · 2014
How to Edit Genes in Human Cells Using an Engineered CRISPR System
This patent describes an engineered CRISPR-Cas9 system for precisely cutting DNA in eukaryotic cells to change how genes work, opening the door for gene editing in complex organisms.
Massachusetts Institute of Technology
US 7657849 · 2010
How the iPhone's Slide-to-Unlock Gesture Works
Apple's 2010 patent describes unlocking a device by dragging a specific graphical image across the touchscreen along a predefined path, a gesture that became iconic with the original iPhone.
Apple Inc
US 4733665 · 1988
How Doctors Implant a Permanent Stent Using a Balloon
This patent describes the method for placing a permanent, expandable wire mesh tube inside a blood vessel or other body tube using a balloon-tipped catheter to widen it and keep it open.
Expandable Grafts Partnership
US 4965188 · 1990
How to Make Many Copies of a DNA Piece with Heat
This patent describes the Polymerase Chain Reaction (PCR) method, a technique to make millions of copies of a specific DNA segment using a heat-resistant enzyme and repeated temperature changes.
Cetus Corp
US 4235871 · 1980
How to Encapsulate Active Materials in Lipid Bubbles Efficiently
This patent describes a method for trapping biologically active substances inside tiny, multi-layered fat bubbles called liposomes, using a specific water-in-oil emulsion and gel-forming process to improve how much material gets captured.
Individual
More to explore
More in AI & Machine Learning
US 10452978 · 2019 · Google LLC
How AI Models Understand Language Using 'Attention'
US 6523026 · 2003 · Huntsman International LLC
How Computers Find Hidden Connections Between Different Fields of Knowledge
US 11615208 · 2023 · Capital One Services LLC
How Cloud Systems Automatically Create and Train AI Data Models
US 10402750 · 2019 · Facebook Inc
How Facebook Uses Deep Learning to Predict What You Might Like
New to patents?
Common Questions
Frequently Asked Questions
What does How to Train AI Models with Fake Data Using Generative Networks cover?
This patent describes a method for training artificial intelligence models using specially generated fake, or 'synthetic,' data created by a generative adversarial network, ensuring the synthetic data is high-quality and safe for training.
Who owns patent US 20230297446?
This patent is owned by Capital One Services.
When does this patent expire?
This patent is expected to expire on May 22, 2043, when the invention enters the public domain.
What problem does this patent solve?
This patent addresses a critical need in AI development: training powerful models without compromising sensitive real-world data. By generating high-quality synthetic data, organizations like Capital One can develop and test AI solutions more rapidly and securely. This approach helps comply with privacy regulations and reduces the risks associated with handling confidential information, enabling innovation in data-sensitive industries.
What does this patent NOT cover?
Does not cover generating synthetic data without using a generative network that specifically includes a decoder network transforming data from a code space to a sample space. (Claim 21)
Same assignee
More from Capital One Services
Patent monitoring