Interactive workflow generation for machine learning lifecycle management

Patent metadata

Field	Value
Publication number	US-11599813-B1
Application number	US-201916584852-A
Country	US
Kind code	B1
Filing date	Sep 26, 2019
Priority date	Sep 26, 2019
Publication date	Mar 7, 2023
Grant date	Mar 7, 2023

Abstract

Methods, systems, and computer-readable media for interactive workflow generation for machine learning lifecycle management are disclosed. A machine learning management system determines one or more prompts associated with use of a machine learning model. Input representing one or more responses to the one or more prompts is received. The one or more responses are provided via a user interface. The machine learning management system determines one or more workflows associated with the machine learning model. The workflow(s) are determined based at least in part on the one or more responses. The workflow(s) comprise a plurality of tasks associated with use of the machine learning model at a plurality of stages of a lifecycle of the model. One or more computing resources are determined, and at least a portion of the workflow(s) is performed using the one or more computing resources.

First claim

What is claimed is:

1. A system, comprising:
one or more pools of computing resources in a multi-tenant provider network; and
one or more computing devices, comprising one or more processors and associated memory, configured to implement a machine learning management system, wherein the machine learning management system is configured to:
determine one or more questions associated with use of a machine learning model, wherein the one or more questions are determined based at least in part on one or more workflow templates, and wherein the one or more questions are displayed via a user interface;
receive input representing one or more answers to the one or more questions, wherein the one or more answers are provided via the user interface;
determine one or more workflows associated with a plurality of stages of a lifecycle of the machine learning model, wherein the one or more workflows are determined based at least in part on the one or more answers, and wherein the one or more workflows comprise one or more tasks associated with training the machine learning model and one or more tasks associated with using the machine learning model to generate inferences;
provision a set of computing resources from the one or more pools of computing resources; and
cause the set of computing resources to perform at least a portion of the one or more tasks associated with training the machine learning model and at least a portion of the one or more tasks associated with using the machine learning model to generate inferences.

2. The system as recited in claim 1 , wherein the one or more answers represent a selection of a particular workflow template, data indicative of one or more inputs for training the machine learning model, and data indicative of one or more inputs for using the machine learning model to generate the inferences.

3. The system as recited in claim 1 , wherein the machine learning management system is further configured to:
add a plurality of versions of the machine learning model to a model registry, wherein the plurality of versions of the machine learning model represent output of a plurality of steps of the one or more workflows; and
based at least in part on the one or more workflows, retrieve a particular version of the plurality of versions of the machine learning model from the model registry, wherein the particular version is used by the one or more tasks associated with using the machine learning model to generate the inferences.

4. The system as recited in claim 1 , wherein the machine learning management system is further configured to:
determine a resource template describing the set of computing resources and representing one or more resource architectures in the multi-tenant provider network; and
merge the resource template into a continuous deployment pipeline in the multi-tenant provider network.

5. A computer-implemented method, comprising:
determining, by a machine learning management system, one or more prompts associated with use of a machine learning model;
receiving input representing one or more responses to the one or more prompts, wherein the one or more responses are provided via a user interface;
determining, by the machine learning management system, one or more workflows associated with the machine learning model, wherein the one or more workflows are determined based at least in part on the one or more responses, and wherein the one or more workflows comprise one or more tasks associated with training the machine learning model, or a plurality of tasks associated with the use of the machine learning model at a plurality of stages of a lifecycle of the machine learning model;
provisioning one or more computing resources; and
causing the one or more computing resources to perform at least a portion of the one or more workflows.

6. The method as recited in claim 5 , wherein provisioning the one or more computing resources further comprises:
provisioning the one or more computing resources from one or more pools of computing resources of a multi-tenant provider network; and
wherein causing the one or more computing resources to perform at least the portion of the one or more workflows further comprises:
causing the one or more computing resources to perform at least a portion of the one or more tasks associated with training the machine learning model.

7. The method as recited in claim 5 , wherein provisioning the one or more computing resources further comprises:
provisioning the one or more computing resources from one or more pools of computing resources of a multi-tenant provider network; and
wherein causing the one or more computing resources to perform at least the portion of the one or more workflows further comprises:
causing the one or more computing resources to perform at least a portion of the plurality of tasks associated with the use of the machine learning model at the plurality of stages of the lifecycle of the machine learning model.

8. The method as recited in claim 5 , wherein determining the one or more computing resources comprises determining a resource template representing one or more resource architectures in a multi-tenant provider network, and wherein the method further comprises:
merging the resource template into a continuous deployment pipeline.

9. The method as recited in claim 5 , wherein the one or more workflows comprise one or more tasks associated with developing a learning algorithm associated with the machine learning model.

10. The method as recited in claim 5 , wherein the one or more workflows comprise one or more tasks associated with determining inputs for use of the machine learning model.

11. The method as recited in claim 5 , wherein the user interface comprises a command-line interface, a graphical user interface, or a voice-enabled interface.

12. The method as recited in claim 5 , further comprising:
receiving additional input representing one or more modifications to the one or more workflows, wherein the one or more modifications are provided via the user interface after the one or more workflows are initiated using the one or more computing resources; and
determining, by the machine learning management system, a modified version of the one or more workflows associated with the machine learning model, wherein the modified version of the one or more workflows is determined based at least in part on the one or more modifications, wherein the modified version of the one or more workflows comprises one or more additional workflow steps or one or more modified workflow steps, and wherein at least a portion of the modified version of the one or more workflows is performed using the one or more computing resources.

13. The method as recited in claim 5 , wherein at least a portion of the one or more workflows is performed using orchestration of a plurality of services of a multi-tenant provider network.

14. The method as recited in claim 5 , further comprising:
adding one or more versions of the machine learning model to a model registry, wherein the model registry maintains a lineage associated with the machine learning model; and
based at least in part on the one or more workflows, retrieving a particular version of the one or more versions of the machine learning model from the model registry and using the particular version in one or more of the tasks of the one or more workflows.

15. One or more non-transitory computer-readable storage media storing program instructions that, when executed on or across one or more processors, perform:
determining, by a machine learning management system, one or more question

Assignees

• Amazon Tech Inc

Inventors

• Yuan Fei
• Huang Shuye

Classifications

• G06F2209/5011

  Pool · CPC title

• G06F9/5005 Primary

  to service a request · CPC title

• G06N5/045 Primary

  Explanation of inference; Explainable artificial intelligence [XAI]; Interpretable artificial intelligence · CPC title

• G06N20/00

  Machine learning · CPC title