# US 2021/0241169 A1 · Performance based switching of a model training process

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

- Number: US 2021/0241169 A1
- Document ID: US20210241169A1
- Status: Published application
- Kind: A1
- Application number: 16/782,713
- Filed: 2020-02-05
- Published: 2021-08-05
- Earliest claimed: 2020-02-05
- Assignees: International Business Machines Corporation
- Applicants: International Business Machines Corporation
- Inventors: Sidharth Gupta; Parijat Dube; Ashish Verma
- Web page: https://patentgrep.com/patent/US20210241169A1

## CPC

- G06N 3/08: Learning methods
- G06N 20/00: Machine learning
- G06F 11/3466: {Performance evaluation by tracing or monitoring}
- G06F 17/18: for evaluating statistical data {, e.g. average values, frequency distributions, probability functions, regression analysis (forecasting specially adapted for a specific administrative, business or logistic context G06Q10/04)}
- G06N 3/0464: Convolutional networks [CNN, ConvNet]
- G06N 3/09: Supervised learning
- G06N 3/094: Adversarial learning
- G06F 11/3688: {for test execution, e.g. scheduling of test suites}
- G06N 3/045: Combinations of networks
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

Systems, computer-implemented methods, and computer program products that can facilitate switching a model training process from a ground truth training phase to an adversarial training phase based on performance of a model trained in the ground truth training phase are provided. According to an embodiment, a system can comprise a memory that stores computer executable components and a processor that executes the computer executable components stored in the memory. The computer executable components can comprise an analysis component that identifies a performance condition of a model trained in a model training process. The computer executable components can further comprise a trainer component that switches the model training process from a ground truth training process to an adversarial training process based on the identified performance condition.

## Claims

### Claim 1

A system, comprising:
a memory that stores computer executable components; and
a processor that executes the computer executable components stored in the memory, wherein the computer executable components comprise:
an analysis component that identifies a performance condition of a model trained in a model training process; and
a trainer component that switches the model training process from a ground truth training process to an adversarial training process based on the identified performance condition.

### Claim 2 (depends on 1)

The system of claim 1, wherein the trainer component trains the model using ground truth training data until the analysis component determines if the model is still evolving by checking if the test accuracy values of the model over a defined number of training epochs are within a defined range of values.

### Claim 3 (depends on 1)

The system of claim 1, wherein the trainer component trains the model using ground truth training data until the analysis component determines that a learning rate reduction of the model is within a defined number of training epochs from a current training epoch.

### Claim 4 (depends on 1)

The system of claim 1, wherein based on identification of the performance condition the trainer component trains the model using adversarial training data until convergence of test accuracy values of the model.

### Claim 5 (depends on 1)

The system of claim 1, wherein the computer executable components further comprise:
a sample generator component that generates adversarial training data to train the model using the adversarial training process based on identification of the performance condition.

### Claim 6 (depends on 1)

The system of claim 1, wherein the analysis component determines a test accuracy value of the model at a current training epoch and computes a running average of test accuracy values of the model over a defined number of training epochs to identify the performance condition.

### Claim 7 (depends on 1)

The system of claim 1, wherein the trainer component switches the model training process from the ground truth training process to the adversarial training process based on the performance condition to facilitate at least one of reduced overall robustness training time, reduced adversarial sample generation time, reduced computational costs of the processor, or improved accuracy of the model trained using ground truth training data and adversarial training data.

### Claim 8

A computer-implemented method, comprising:
identifying, by a system operatively coupled to a processor, a performance condition of a model trained in a model training process; and
switching, by the system, the model training process from a ground truth training process to an adversarial training process based on the identified performance condition.

### Claim 9 (depends on 8)

The computer-implemented method of claim 8, further comprising:
training, by the system, the model using ground truth training data until test accuracy values of the model over a defined number of training epochs are within a defined range of values.

### Claim 10 (depends on 8)

The computer-implemented method of claim 8, further comprising:
training, by the system, the model using ground truth training data until a learning rate reduction of the model is within a defined number of training epochs from a current training epoch.

### Claim 11 (depends on 8)

The computer-implemented method of claim 8, further comprising:
training, by the system, the model, based on identification of the performance condition, using adversarial training data until convergence of test accuracy values of the model.

### Claim 12 (depends on 8)

The computer-implemented method of claim 8, further comprising:
generating, by the system, adversarial training data to train the model using the adversarial training process based on identification of the performance condition.

### Claim 13 (depends on 8)

The computer-implemented method of claim 8, further comprising:
determining, by the system, a test accuracy value of the model at a current training epoch; and
computing, by the system, a running average of test accuracy values of the model over a defined number of training epochs to identify the performance condition.

### Claim 14 (depends on 8)

The computer-implemented method of claim 8, further comprising:
switching, by the system, the model training process from the ground truth training process to the adversarial training process based on the performance condition to facilitate at least one of reduced overall robustness training time, reduced adversarial sample generation time, reduced computational costs of the processor, or improved accuracy of the model trained using ground truth training data and adversarial training data.

### Claim 15

A computer program product facilitating a process to switch a model training process from a ground truth training phase to an adversarial training phase based on performance of a model trained in the ground truth training phase, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a processor to cause the processor to:
identify, by the processor, a performance condition of a model trained in a model training process; and
switch, by the processor, the model training process from a ground truth training process to an adversarial training process based on the identified performance condition.

### Claim 16 (depends on 15)

The computer program product of claim 15, wherein the program instructions are further executable by the processor to cause the processor to:
train, by the processor, the model using ground truth training data until test accuracy values of the model over a defined number of training epochs are within a defined range of values.

### Claim 17 (depends on 15)

The computer program product of claim 15, wherein the program instructions are further executable by the processor to cause the processor to:
train, by the processor, the model using ground truth training data until a learning rate reduction of the model is within a defined number of training epochs from a current training epoch.

### Claim 18 (depends on 15)

The computer program product of claim 15, wherein the program instructions are further executable by the processor to cause the processor to:
train, by the processor, the model, based on identification of the performance condition, using adversarial training data until convergence of test accuracy values of the model.

### Claim 19 (depends on 15)

The computer program product of claim 15, wherein the program instructions are further executable by the processor to cause the processor to:
generate, by the processor, adversarial training data to train the model using the adversarial training process based on identification of the performance condition.

### Claim 20 (depends on 15)

The computer program product of claim 15, wherein the program instructions are further executable by the processor to cause the processor to:
determine, by the processor, a test accuracy value of the model at a current training epoch; and
compute, by the processor, a running average of test accuracy values of the model over a defined number of training epochs to identify the performance condition.
