IP Library Granted Patent US 12,725,075
Granted Patent B2
US 12,725,075 · App. 17/482,068 · Granted Sep 1, 2026

Training data generation via reinforcement learning fault-injection

Inventors: Jinho Hwang (Dobbs Ferry, NY); Larisa Shwartz (Greenwich, CT); Jesus Maria Rios Aliaga (Philadelphia, PA); Frank Bagehorn (Dottikon, CH); Stephen James Hussey (Orange, CA)
Assignee: INTERNATIONAL BUSINESS MACHINES CORPORATION
G06N20/00G06F11/3476G06F18/217
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12,725,075
App. No.
17/482,068
Granted
Sep 1, 2026
Kind
B2
Abstract

Systems/techniques for generating training data via reinforcement learning fault-injection are provided. A system can access a computing application. In various aspects, the system can train one or more machine learning models based on responses of the computing application to iterative fault-injections determined via reinforcement learning. More specifically, the system can: inject a first fault into the computing application; record a resultant dataset outputted by the computing application in response to the first fault; train the one or more machine learning models on the resultant dataset and the first fault; compute a reinforcement learning reward based on performance metrics of the one or more machine learning models and based on a quantity of the resultant dataset; update, via execution of a reinforcement learning algorithm, the fault-injection policy based on the reinforcement learning reward; and inject a second fault into the computing application, based on the updated fault-injection policy.

Claims (47)

1 . A system, comprising:

a processor that executes computer-executable components stored in a computer-readable memory, the computer-executable components comprising:

a transceiver component that accesses a computing application;

a training component that trains one or more machine learning models based on responses of the computing application to iterative fault-injections that are determined via reinforcement learning, wherein at least one of the fault-injections comprises an error, and wherein a size of error data and performance quality of the one or more machine learning models after being trained on the error data collectively dictate a reward of the reinforcement learning;

a reward component that computes a reinforcement learning reward based on one or more performance metrics of the one or more machine learning models after the training; and

an update component that updates, via execution of a reinforcement learning algorithm and employing backpropagation, the fault-injection policy based on the reinforcement learning reward.

2 . The system of claim 1 , wherein the computer-executable components further comprise:

a fault-injection component that injects a first fault into the computing application, based on a fault-injection policy.

3 . The system of claim 2 , wherein the computer-executable components further comprise:

a logging component that records a resultant dataset outputted by the computing application in response to the first fault.

4 . The system of claim 3 , wherein the training component trains the one or more machine learning models on the resultant dataset and the first fault.

5 . The system of claim 4 , wherein the computer-executable:

a reward component

also computes a quantity of the resultant dataset.

6 . The system of claim 1 , wherein the fault-injection component injects a second fault into the computing application, based on the updated fault-injection policy.

7 . A computer-implemented method, comprising:

accessing, by a device operatively coupled to a processor, a computing application;

training, by the device, one or more machine learning models based on responses of the computing application to iterative fault-injections that are determined via reinforcement learning, wherein at least one of the fault-injections comprises an error, and wherein concurrent consideration of a size of error data and a performance quality of the one or more machine learning models after being trained on the error data dictate a reward of the reinforcement learning;

computing, by the device, a reinforcement learning reward based on one or more performance metrics of the one or more machine learning models after the training; and

updating, by the device, via execution of a reinforcement learning algorithm and employing backpropagation, the fault-injection policy based on the reinforcement learning reward.

8 . The computer-implemented method of claim 7 , wherein the training the one or more machine learning models based on responses of the computing application to iterative fault-injections includes:

injecting, by the device, a first fault into the computing application, based on a fault-injection policy.

9 . The computer-implemented method of claim 8 , wherein the training the one or more machine learning models based on responses of the computing application to iterative fault-injections further includes:

recording, by the device, a resultant dataset outputted by the computing application in response to the first fault.

10 . The computer-implemented method of claim 9 , wherein the training the one or more machine learning models based on responses of the computing application to iterative fault-injections further includes:

training, by the device, the one or more machine learning models on the resultant dataset and the first fault.

11 . The computer-implemented method of claim 10 , wherein the training the one or more machine learning models based on responses of the computing application to iterative fault-injections further includes:

computing, by the device, a quantity of the resultant dataset.

12 . The computer-implemented method of claim 11 , wherein the training the one or more machine learning models based on responses of the computing application to iterative fault-injections further includes:

updating, by the device and via execution of a reinforcement learning algorithm, the fault-injection policy based on the reinforcement learning reward.

13 . The computer-implemented method of claim 12 , wherein the training the one or more machine learning models based on responses of the computing application to iterative fault-injections further includes:

injecting, by the device, a second fault into the computing application, based on the updated fault-injection policy.

14 . A computer program product for facilitating training data generation via reinforcement learning fault-injection, the computer program product comprising a computer-readable memory having program instructions embodied therewith, the program instructions executable by a processor to cause the processor to:

access, by the processor, a computing application;

train, by the processor, one or more machine learning models based on responses of the computing application to iterative fault-injections that are determined via reinforcement learning wherein at least one of the fault-injections comprises an error, and wherein a size of error data and a performance quality of the one or more machine learning models after being trained on the error data collectively dictate comprises a reward of the reinforcement learning;

compute, by the processor, a reinforcement learning reward based on one or more performance metrics of the one or more machine learning models after the training; and

update, by the processor, via execution of a reinforcement learning algorithm and employing backpropagation, the fault-injection policy based on the reinforcement learning reward.

15 . The computer program product of claim 14 , wherein the processor trains the one or more machine learning models based on responses of the computing application to iterative fault-injections by:

injecting, by the processor, a first fault into the computing application, based on a fault-injection policy.

16 . The computer program product of claim 15 , wherein the processor trains the one or more machine learning models based on responses of the computing application to iterative fault-injections by:

recording, by the processor, a resultant dataset outputted by the computing application in response to the first fault.

17 . The computer program product of claim 16 , wherein the processor trains the one or more machine learning models based on responses of the computing application to iterative fault-injections by:

training, by the processor, the one or more machine learning models on the resultant dataset and the first fault.

18 . The computer program product of claim 17 , wherein the processor trains the one or more machine learning models based on responses of the computing application to iterative fault-injections by:

computing, by the processor, a quantity of the resultant dataset.

19 . The computer program product of claim 18 , wherein the processor trains the one or more machine learning models based on responses of the computing application to iterative fault-injections by:

updating, by the processor and via execution of a reinforcement learning algorithm, the fault-injection policy based on the reinforcement learning reward.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Sep 22, 2021
From: HWANG, JINHO; SHWARTZ, LARISA; RIOS ALIAGA, JESUS MARIA; BAGEHORN, FRANK; HUSSEY, STEPHEN JAMES
To: INTERNATIONAL BUSINESS MACHINES CORPORATION
Reel/Frame 057566/0008 →
Continuity (1)
Related Publication 20230087837A1 · Mar 23, 2023
References Cited (22)
US 7225200B2 · Chickering et al. · 2007 [cited by applicant]
US 10986013B1 · Theimer · 2021 [cited by examiner]
US 20170243140A1 · Achin et al. · 2017 [cited by applicant]
US 20190220745A1 · Huang et al. · 2019 [cited by applicant]
US 20190220776A1 · Huang et al. · 2019 [cited by applicant]
US 20190370473A1 · Matrosov et al. · 2019 [cited by applicant]
US 20190384699A1 · Arbon · 2019 [cited by examiner]
CN 109495297A · 2019 [cited by examiner]
CN 110389892A · 2019 [cited by applicant]
CN 110427275A · 2019 [cited by applicant]
CN 111209131A · 2020 [cited by applicant]
CN 111240304A · 2020 [cited by applicant]
CN 111597122A · 2020 [cited by applicant]
CN 112214922A · 2021 [cited by examiner]
CN 115934455A · 2023 [cited by applicant]
JP 2021128538A · 2021 [cited by examiner]
JP 2023046293A · 2023 [cited by applicant]
“Faults in Deep Reinforcement Learning Programs: A taxonomy and a Detection Approach”, Nikanjam et al, Jan. 2, 2021 (Year: 2021). [cited by examiner]
“Exploring Fault Parameter Space Using Reinforcement Learning-based Fault Injection”, Moradi et al , 2020. [cited by examiner]
“Using Machine Learning Techniques to Evaluate Multicore Soft Error Reliability”, Rocha De Rosa et al, 2019 (Year: 2019). [cited by examiner]
Mell et al., “The NIST Definition of Cloud Computing,” Recommendations of the National Institute of Standards and Technology, NIST Special Publication 800-145, Sep. 2011, 7 pages. [cited by applicant]
Japan Patent Office, “Notice of Reasons for Refusal” Jan. 20, 2026, 08 Pages, JP Application No. 2022-147140. [cited by applicant]