# US 12,223,281 B2 · Generating content using a generative model without relying on selected training examples

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

- Number: US 12,223,281 B2
- Document ID: US12223281B2
- Status: Granted
- Kind: B2
- Application number: 18/387,701
- Filed: 2023-11-07
- Published: 2024-08-15
- Granted: 2025-02-11
- Earliest claimed: 2023-02-10
- Assignees: Bria Artificial Intelligence Ltd.
- Applicants: Bria Artificial Intelligence Ltd.
- Inventors: Yair Adato; Efrat Taig; Nimrod Sarid; Ron Mokady; Eyal Gutflaish
- Web page: https://patentgrep.com/patent/US12223281B2

## CPC

- G06F 40/30: Semantic analysis
- G06N 3/08: Learning methods
- G06F 40/279: Recognition of textual entities
- G06F 40/40: Processing or translation of natural language (natural language analysis G06F40/20; semantic analysis G06F40/30)
- G06N 3/044: Recurrent networks, e.g. Hopfield networks
- G06N 3/045: Combinations of networks
- G06N 3/084: Backpropagation, e.g. using gradient descent
- G06N 20/00: Machine learning
- G06T 7/194: involving foreground-background segmentation
- G06T 7/70: Determining position or orientation of objects or cameras (camera calibration G06T7/80)
- G06T 11/10: Texturing; Colouring; Generation of textures or colours (retouching, inpainting or scratch removal G06T5/77)
- G06V 10/764: using classification, e.g. of video objects
- G06V 10/774: Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
- G10L 15/063: {Training}
- G10L 15/18: using natural language modelling
- H04N 5/272: Means for inserting a foreground image in a background image, i.e. inlay, outlay
- G06F 40/211: Syntactic parsing, e.g. based on context-free grammar [CFG] or unification grammars
- G06F 40/284: Lexical analysis, e.g. tokenisation or collocates
- G06T 2207/30196: Human being; Person
- G10L 2015/0631: {Creating reference templates; Clustering}
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G10L: SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- H04N: PICTORIAL COMMUNICATION, e.g. TELEVISION

## Abstract

Systems, methods and non-transitory computer readable media for generating content using a generative model without relying on selected training examples are provided. An input indicative of a desire to generate a new content using a generative model may be received. The generative model may be a result of training a machine learning model using a plurality of training examples. Each training example of the plurality of training examples may be associated with a respective content. Further, an indication of a particular subgroup of at least one but not all of the plurality of training examples may be obtained. Based on the indication, the input and the generative model may be used to generate the new content, abstaining from basing the generation of the new content on any training example included in the particular subgroup. The new content may be provided.

## Claims

### Claim 1

A non-transitory computer readable medium storing a software program comprising data and computer implementable instructions that when executed by at least one processor cause the at least one processor to perform operations for generating content using a generative model, the operations comprising
receiving an input indicative of a desire to generate a new content using a generative model, the generative model is a result of training a machine learning model using all training examples of a plurality of training examples, each training example of the plurality of training examples is associated with a respective content, the generative model includes a plurality of artificial neurons;
obtaining an indication of a particular subgroup of at least one but not all of the plurality of training examples;
identifying a specific subgroup of the plurality of artificial neurons associated with the particular subgroup of at least one but not all of the plurality of training examples;
based on the indication, using the input and the generative model while disregarding the artificial neurons included in the specific subgroup to generate the new content, abstaining, after the training of the machine learning model is completed, from basing the generation of the new content on any training example included in the particular subgroup; and
providing the new content.

### Claim 2 (depends on 1)

The non-transitory computer readable medium of claim 1, wherein the new content is at least one of a visual content, an audio content or a textual content.

### Claim 3 (depends on 1)

The non-transitory computer readable medium of claim 1, wherein the particular subgroup is selected based on a geographical region.

### Claim 4 (depends on 1)

The non-transitory computer readable medium of claim 1, wherein the particular subgroup is selected based on a regulatory requirement.

### Claim 5 (depends on 1)

The non-transitory computer readable medium of claim 1, wherein the particular subgroup is selected based on a source associated with the contents associated with the particular subgroup.

### Claim 6 (depends on 1)

The non-transitory computer readable medium of claim 1, wherein the operations further comprise analyzing the input to select the particular subgroup.

### Claim 7 (depends on 1)

The non-transitory computer readable medium of claim 1, wherein the operations further comprise:
calculating a convolution of at least part of the content associated with a particular training example of the plurality of training examples to obtain a numerical result value; and
determining whether to include the particular training example in the particular subgroup based on the numerical result value.

### Claim 8 (depends on 1)

The non-transitory computer readable medium of claim 1, wherein the operations further comprise:
for each training example of the plurality of training examples, analyzing the respective content associated with the training example to identify a respective mathematical object corresponding to the respective training example in a mathematical space;
based on the identified mathematical objects, identifying a region in the mathematical space;
using the input to select a specific mathematical object in the mathematical space outside the identified region; and
using the specific mathematical object to generate the new content.

### Claim 9 (depends on 1)

The non-transitory computer readable medium of claim 1, wherein the operations further comprise:
using the input and the generative model to generate a plurality of alternative contents;
for each content of the plurality of alternative contents, determining a likelihood that the generation of the content is based on any training example included in the particular subgroup; and
selecting the new content of the plurality of alternative contents based on the determined likelihoods.

### Claim 10 (depends on 1)

The non-transitory computer readable medium of claim 1, wherein the operations further comprise:
using the input and the generative model to generate a preliminary content; and
based on the indication, analyzing the preliminary content to modify the preliminary content.

### Claim 11 (depends on 1)

The non-transitory computer readable medium of claim 1, wherein the generative model includes a plurality of weights, and the operations further comprise:
identifying a specific subgroup of the plurality of weights associated with the particular subgroup of at least one but not all of the plurality of training examples; and
abstaining from basing the generation of the new content on any one of the specific subgroup of the plurality of weights.

### Claim 12 (depends on 1)

The non-transitory computer readable medium of claim 1, wherein the generative model includes a memory unit that stores a plurality of patterns, and the operations further comprise:
identifying a specific subgroup of the plurality of patterns associated with the particular subgroup of at least one but not all of the plurality of training examples; and
abstaining from basing the generation of the new content on any one of the specific subgroup of the plurality of patterns.

### Claim 13 (depends on 1)

The non-transitory computer readable medium of claim 1, wherein the input includes a textual input indicative of the desire to generate the new content using the generative model.

### Claim 14 (depends on 13)

The non-transitory computer readable medium of claim 13, wherein the textual input includes a noun and an adjective adjacent to the noun, wherein for each training example of the particular subgroup, the respective content associated with the training example includes a respective object of a category of objects selected based on the noun wherein at least one aspect of the respective object matches the adjective.

### Claim 15 (depends on 13)

The non-transitory computer readable medium of claim 13, wherein the textual input includes a verb and an adverb adjacent to the verb, wherein for each training example of the particular subgroup, the respective content associated with the training example includes a respective event of a category of events selected based on the verb wherein at least one aspect of the respective event matches the adverb.

### Claim 16 (depends on 13)

The non-transitory computer readable medium of claim 13, wherein the textual input includes a preposition, wherein for each training example of the particular subgroup, the respective content associated with the training example includes a respective pair of objects wherein a spatial relation between the pair of objects in the respective content matches the preposition.

### Claim 17 (depends on 1)

The non-transitory computer readable medium of claim 1, wherein the input includes input image data.

### Claim 18 (depends on 1)

The non-transitory computer readable medium of claim 1, wherein each training example of the plurality of training examples includes a visual content, and the operations further comprise, and for each artificial neuron of the plurality of artificial neurons using the artificial neuron and properties of the visual contents associated with the plurality of training examples to attribute the artificial neuron to the particular subgroup, thereby determining whether to include the artificial neuron in the specific subgroup.

### Claim 19

A system for generating content using a generative model without relying on selected training examples, the system comprising:
at least one processor configured to perform the operations of:
receiving an input indicative of a desire to generate a new content using a generative model, the generative model is a result of training a machine learning model using all training examples of a plurality of training examples, each training example of the plurality of training examples is associated with a respective content, the generative model includes a plurality of artificial neurons;
obtaining an indication of a particular subgroup of at least one but not all of the plurality of training examples;
identifying a specific subgroup of the plurality of artificial neurons associated with the particular subgroup of at least one but not all of the plurality of training examples;
based on the indication, using the input and the generative model while disregarding the artificial neurons included in the specific subgroup to generate the new content, abstaining, after the training of the machine learning model is completed, from basing the generation of the new content on any training example included in the particular subgroup; and
providing the new content.

### Claim 20

A method for generating content using a generative model without relying on selected training examples, the method comprising:
receiving an input indicative of a desire to generate a new content using a generative model, the generative model is a result of training a machine learning model using all training examples of a plurality of training examples, each training example of the plurality of training examples is associated with a respective content, the generative model includes a plurality of artificial neurons;
obtaining an indication of a particular subgroup of at least one but not all of the plurality of training examples;
identifying a specific subgroup of the plurality of artificial neurons associated with the particular subgroup of at least one but not all of the plurality of training examples;
based on the indication, using the input and the generative model while disregarding the artificial neurons included in the specific subgroup to generate the new content, abstaining, after the training of the machine learning model is completed, from basing the generation of the new content on any training example included in the particular subgroup; and
providing the new content.