# US 2021/0173736 A1 · Detection of failure conditions and restoration of deployed models in a computing environment

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

- Number: US 2021/0173736 A1
- Document ID: US20210173736A1
- Status: Published application
- Kind: A1
- Application number: 16/704,083
- Filed: 2019-12-05
- Published: 2021-06-10
- Earliest claimed: 2019-12-05
- Assignees: International Business Machines Corporation
- Applicants: International Business Machines Corporation
- Inventors: Evelyn Duesterwald; Punleuk Oum; Gaodan Fang; Debashish Saha; Anupama Murthi; Waldemar Hummer
- Web page: https://patentgrep.com/patent/US20210173736A1

## CPC

- G06F 11/0778: {Dumping, i.e. gathering error/state information after a fault for later diagnosis}
- G06F 11/0793: {Remedial or corrective actions (recovery from an exception in an instruction pipeline G06F9/3861; by retry G06F11/1402; for recovering from a failure of a protocol instance or entity H04L69/40)}
- G06F 11/0751: {Error or fault detection not based on redundancy (power supply failures G06F1/30; network fault management H04L41/06)}
- G06F 11/0772: {Means for error signaling, e.g. using interrupts, exception flags, dedicated error registers}
- G06F 11/079: {Root cause analysis, i.e. error or fault diagnosis (in a hardware test environment G06F11/22; in a software test environment G06F11/36)}
- G06N 20/00: Machine learning
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

A computer-implemented method includes obtaining data associated with execution of a model deployed in a computing environment. At least a portion of the obtained data are analyzed to detect one or more failure conditions associated with the model. One or more restoration operations are executed to generate one or more restoration results to address one or more detected failure conditions. At least a portion of the one or more restoration results is sent to the computing environment in which the model is deployed.

## Claims

### Claim 1

An apparatus comprising:
at least one processor;
at least one memory including computer program code;
the at least one memory and the computer program code being configured to, with the at least one processor, cause the apparatus at least to:
obtain data associated with execution of a model deployed in a computing environment;
analyze at least a portion of the obtained data to detect one or more failure conditions associated with the model;
execute one or more restoration operations to generate one or more restoration results to address one or more detected failure conditions; and
send at least a portion of the one or more restoration results to the computing environment in which the model is deployed.

### Claim 2 (depends on 1)

The apparatus of claim 1, wherein the at least one memory and the computer program code are further configured to, with the at least one processor, cause the apparatus at least to apply at least a portion of the one or more restoration results to the model.

### Claim 3 (depends on 1)

The apparatus of claim 1, wherein the at least one memory and the computer program code are further configured to, with the at least one processor, cause the apparatus at least to validate the one or more restoration results prior to sending at least a portion of one or more restoration results to the computing environment in which the model is deployed.

### Claim 4 (depends on 1)

The apparatus of claim 1, wherein the one or more restoration results comprise an updated version of the model.

### Claim 5 (depends on 1)

The apparatus of claim 1, wherein the one or more restoration results comprise an updated training data set for the model.

### Claim 6 (depends on 1)

The apparatus of claim 1, wherein the obtained data associated with execution of the model deployed in the computing environment comprises usage log data accumulated during a model runtime period.

### Claim 7 (depends on 1)

The apparatus of claim 1, wherein the at least one memory and the computer program code are further configured to, with the at least one processor, cause the apparatus at least to analyze the portion of the obtained data to detect the one or more failure conditions associated with the model using one or more failure condition components.

### Claim 8 (depends on 7)

The apparatus of claim 7, wherein the one or more failure condition components comprise one or more of a bias detection component, a drift detection component and a security threat level component.

### Claim 9 (depends on 7)

The apparatus of claim 7, wherein the at least one memory and the computer program code are further configured to, with the at least one processor, cause the apparatus at least to execute the one or more restoration operations to generate the one or more restoration results to address the one or more detected failure conditions using one or more restoration pipelines respectively operatively coupled to the one or more failure condition components.

### Claim 10 (depends on 9)

The apparatus of claim 9, wherein the one or more restoration pipelines comprise one or more of a bias removal pipeline, a drift-responsive re-training pipeline and an adversarial training pipeline.

### Claim 11 (depends on 9)

The apparatus of claim 9, wherein when multiple ones of the failure condition components trigger execution of multiple ones of the restoration pipelines, the at least one memory and the computer program code are further configured to, with the at least one processor, cause the apparatus at least to evaluate multiple restoration results generated by the multiple ones of the restoration pipelines and at least one of the evaluated restoration results is sent to the computing environment in which the model is deployed.

### Claim 12 (depends on 1)

The apparatus of claim 1, wherein the at least one memory and the computer program code are further configured to, with the at least one processor, cause the apparatus at least to combine execution of the one or more restoration operations into a dynamically-constructed pipeline to generate the one or more restoration results to address one or more detected failure conditions.

### Claim 13 (depends on 1)

The apparatus of claim 1, wherein the computing environment in which the model is deployed is a cloud computing environment.

### Claim 14

A computer-implemented method comprising:
obtaining data associated with execution of a model deployed in a computing environment;
analyzing at least a portion of the obtained data to detect one or more failure conditions associated with the model;
executing one or more restoration operations to generate one or more restoration results to address one or more detected failure conditions; and
sending at least a portion of the one or more restoration results to the computing environment in which the model is deployed.

### Claim 15 (depends on 14)

The method of claim 14, further comprising applying at least a portion of the one or more restoration results to the model.

### Claim 16 (depends on 14)

An article of manufacture comprising a non-transitory computer-readable storage medium having embodied therein executable program code that when executed by a processor causes the processor to perform the steps of claim 14.

### Claim 17

A computer-implemented method comprising:
instantiating a detection and restoration framework; and
coupling the instantiated detection and restoration framework to a computing environment in which a model is deployed;
wherein the instantiated detection and restoration framework is configured to obtain data associated with execution of the model, analyze at least a portion of the obtained data to detect one or more failure conditions associated with the model, execute one or more restoration operations to generate one or more restoration results to address one or more detected failure conditions, and send at least a portion of the one or more restoration results to the computing environment in which the model is deployed.

### Claim 18 (depends on 17)

The method of claim 17, wherein the instantiation step further comprises:
selectively obtaining at least one of a set of one or more pre-built failure detection components and a set of one or more custom-built failure detection components; and
configuring the selectively obtained failure detection components.

### Claim 19 (depends on 18)

The method of claim 18, wherein the instantiation step further comprises:
selectively obtaining at least one of a set of one or more restoration pipelines and a set of one or more custom-built restoration pipelines; and
configuring the selectively obtained restoration pipelines.

### Claim 20 (depends on 17)

An article of manufacture comprising a non-transitory computer-readable storage medium having embodied therein executable program code that when executed by a processor causes the processor to perform the steps of claim 17.
