# US 2026/0037637 A1 · Training and using multi-task predictive risk models in software applications

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

- Number: US 2026/0037637 A1
- Document ID: US20260037637A1
- Status: Published application
- Kind: A1
- Application number: 18/790,505
- Filed: 2024-07-31
- Published: 2026-02-05
- Earliest claimed: 2024-07-31
- Assignees: Intuit Inc.
- Applicants: Intuit Inc.
- Inventors: Nazanin Zaker HABIBABADI; Lihua HAN; Ariel NIKZAD; Xue HAN
- Web page: https://patentgrep.com/patent/US20260037637A1

## CPC

- G06F 21/577: {Assessing vulnerabilities and evaluating computer system security}
- G06N 3/045: Combinations of networks
- G06N 3/0464: Convolutional networks [CNN, ConvNet]
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

Certain aspects of the present disclosure provide techniques for executing operations in a software application based on an output of a multi-task machine learning model. An example method generally includes receiving, from a user of a computing system, a request to execute a task in the computing system. Based on receiving the request, an input feature vector is generated based on historical data associated with the user. A plurality of predictive classifications is requested from a multi-output machine learning model based on the input feature vector. Generally, each respective predictive classification of the plurality of predictive classifications is associated with a respective likelihood of whether a specified event will occur for the user. The task is executed based on the plurality of predictive classifications.

## Claims

### Claim 1

A processor-implemented method, comprising:
receiving, from a user of a computing system, a request to execute a task in the computing system;
generating, based on the request, an input feature vector based on historical data associated with the user;
requesting, from a multi-output machine learning model, a plurality of predictive classifications based on the input feature vector, wherein each respective predictive classification of the plurality of predictive classifications is associated with a respective likelihood of whether a specified event will occur for the user; and
executing the task based on the plurality of predictive classifications.

### Claim 2 (depends on 1)

The method of claim 1, wherein the plurality of predictive classifications is associated with a likelihood of a same event occurring over different time windows.

### Claim 3 (depends on 1)

The method of claim 1, wherein the plurality of predictive classifications is associated with different types of events.

### Claim 4 (depends on 1)

The method of claim 1, further comprising reducing the plurality of predictive classifications into a single classification score, wherein the task is executed based on the single classification score.

### Claim 5 (depends on 4)

The method of claim 4, wherein the single classification score identifies, in a user segmentation model, a risk segment in which the user lies.

### Claim 6 (depends on 1)

The method of claim 1, wherein the input feature vector is generated based on a historical transaction history data set for the user.

### Claim 7 (depends on 1)

The method of claim 1, wherein each classification of the plurality of predictive classifications is associated with a range of external risk scores.

### Claim 8 (depends on 1)

The method of claim 1, wherein the multi-output machine learning model comprises a regularizing gradient boosting model with local explainability values associated with one or more features in the input feature vector.

### Claim 9 (depends on 1)

The method of claim 1, wherein the multi-output machine learning model comprises a convolutional neural network including a set of layers common to the plurality of predictive classifications and a plurality of classification task-specific layers, each respective layer of the plurality of classification task-specific layers being associated with a respective classification task.

### Claim 10 (depends on 1)

The method of claim 1, wherein the task comprises transferring assets from the user to another user of a software application.

### Claim 11 (depends on 1)

The method of claim 1, wherein the task comprises accessing one or more access-controlled digital assets in an electronic system.

### Claim 12 (depends on 1)

The method of claim 1, wherein the plurality of predictive classifications indicates whether the request is a fraudulent or legitimate request to perform the task in the computing system.

### Claim 13

A processor-implemented method, comprising:
extracting, from a transaction history data set for a plurality of users of a software application, a plurality of features for each user of the plurality of users having records in the transaction history data set;
generating a training data set based on the extracted plurality of features for each user of the plurality of users; and
training a multi-output machine learning model to generate a plurality of predictive classifications for an input of the plurality of features based on the training data set, wherein each respective predictive classification of the plurality of predictive classifications is associated with a respective likelihood of whether a specified event will occur for a user of a computing system.

### Claim 14 (depends on 13)

The method of claim 13, wherein the multi-output machine learning model comprises a regularizing gradient boosting model with local explainability values associated with one or more features in the input.

### Claim 15 (depends on 13)

The method of claim 13, wherein the multi-output machine learning model comprises a convolutional neural network including a set of layers common to the plurality of predictive classifications and a plurality of classification task-specific layers, each respective layer of the plurality of classification task-specific layers being associated with a respective classification task.

### Claim 16 (depends on 13)

The method of claim 13, wherein the plurality of predictive classifications is associated with a likelihood of a same event occurring over different time windows.

### Claim 17 (depends on 13)

The method of claim 13, wherein the plurality of predictive classifications is associated with different types of events.

### Claim 18

A processing system, comprising:
at least one memory having executable instructions stored thereon; and
one or more processors configured to execute the executable instructions to cause the processing system to:
receive, from a user of a computing system, a request to execution of a task in the computing system;
generate an input feature vector based on historical data associated with the user;
request, from a multi-output machine learning model, a plurality of predictive classifications, wherein each respective predictive classification is associated with a respective likelihood of whether a specified event will occur for the user; and
execute the task based on the plurality of predictive classifications.

### Claim 19 (depends on 18)

The processing system of claim 18, wherein the plurality of predictive classifications comprises a plurality of predictive classifications associated with a likelihood of a same event occurring over different time windows.

### Claim 20 (depends on 18)

The processing system of claim 18, wherein the plurality of predictive classifications comprises a plurality of predictive classifications associated with different types of events.
