Network-accessible machine learning model training and hosting system

Patent metadata

Field	Value
Publication number	US-11977958-B2
Application number	US-201715821585-A
Country	US
Kind code	B2
Filing date	Nov 22, 2017
Priority date	Nov 22, 2017
Publication date	May 7, 2024
Grant date	May 7, 2024

Abstract

A network-accessible machine learning service is provided herein. For example, the network-accessible machine learning service provider can operate one or more physical computing devices accessible to user devices via a network. These physical computing device(s) can host virtual machine instances that are configured to train machine learning models using training data referenced by a user device. These physical computing device(s) can further host virtual machine instances that are configured to execute trained machine learning models in response to user-provided inputs, generating outputs that are stored and/or transmitted to user devices via the network.

First claim

What is claimed is:

1. A system comprising:
a training model data store; and
a first computing device to host a first virtual machine instance, the first computing device comprising computer-executable instructions that, when executed, cause the first computing device to:
obtain, from a user device, a training request, wherein the training request comprises an indication of a storage location of a container image, an indicator of training data, and one or more first hyperparameter values;
initialize a machine learning (ML) training container in the first virtual machine instance, wherein the ML training container is formed from the container image;
cause the first virtual machine instance to execute code stored within the ML training container according to the one or more first hyperparameter values, wherein execution of the code causes the first virtual machine instance to train a machine learning model using the training data and to generate model data that represents characteristics of the machine learning model;
store the model data in the training model data store;

obtain a modification request to modify the machine learning model being trained, wherein the modification request comprises an indicator of a second container image;
cause the first virtual machine instance to stop execution of the code;
replace the ML training container with a second ML training container formed using the second container image; and
cause the first virtual machine instance to execute second code stored within the second ML training container.

2. The system of claim 1 , wherein execution of the second code causes the first machine instance to re-train the machine learning model and to generate second model data.

3. The system of claim 1 , wherein the computer-executable instructions, when executed, further cause the first computing device to:
obtain one or more second hyperparameter values; and
cause the first virtual machine instance to execute the code stored within the ML training container according to the one or more second hyperparameter values instead of the one or more first hyperparameter values.

4. The system of claim 1 , wherein the computer-executable instructions, when executed, further cause the first computing device to:
obtain evaluation data, wherein the evaluation data comprises input data and known results;
execute the machine learning model defined by the model data using the input data as inputs to generate model output data;
compare the model output data with the known results to determine a quality metric of the machine learning model; and
store the quality metric.

5. A computer-implemented method comprising:
receiving, from a user device over a network, a training request, wherein the training request comprises an indication of a storage location of a container image and an indicator of training data;
initializing a machine learning (ML) training container in a first virtual machine instance hosted by a first computing device, wherein the ML training container is formed from the container image;
causing the first virtual machine instance to execute code stored within the ML training container, wherein execution of the code causes the first virtual machine instance to train a machine learning model using the training data and to generate model data that represents characteristics of the machine learning model;
receiving a modification request to modify the machine learning model being trained, wherein the modification request comprises an indicator of a second container image;
causing the first virtual machine instance to stop execution of the code;
initializing a second ML training container in the first virtual machine instance, wherein the second ML training container is formed using the second container image; and
causing the first virtual machine instance to execute second code stored within the second ML training container.

6. The computer-implemented method of claim 5 , wherein execution of the second code causes the first machine instance to re-train the machine learning model and to generate second model data.

7. The computer-implemented method of claim 5 , wherein the computer-implemented method further comprises:
obtaining evaluation data, wherein the evaluation data comprises input data and known results;
executing the machine learning model defined by the model data using the input data as inputs to generate model output data; and
comparing the model output data with the known results to determine a quality metric of the machine learning model.

8. comprisingA computer-implemented method comprising:
receiving, from a user device over a network, a training request, wherein the training request comprises an indication of a storage location of a container image and an indicator of training data;
initializing a machine learning (ML) training container in a first virtual machine instance hosted by a first computing device, wherein the ML training container is formed from the container image;
causing the first virtual machine instance to execute code stored within the ML training container, wherein execution of the code causes the first virtual machine instance to train a machine learning model using the training data and to generate model data that represents characteristics of the machine learning model;
initializing a second ML training container in the first virtual machine instance, wherein the second ML training container is formed from the container image; and
causing the first virtual machine instance to execute second code stored within the second ML training container in parallel with the execution of the code stored within the ML training container, wherein execution of the second code causes the first machine instance to generate second model data, and wherein a combination of the model data and the second model data defines characteristics of a trained version of the machine learning model.

9. A computer-implemented method comprising:
receiving, from a user device over a network, a training request, wherein the training request comprises an indication of a storage location of a container image and an indicator of training data;
initializing a machine learning (ML) training container in a first virtual machine instance hosted by a first computing device, wherein the ML training container is formed from the container image;
causing the first virtual machine instance to execute code stored within the ML training container, wherein execution of the code causes the first virtual machine instance to train a machine learning model using the training data and to generate model data that represents characteristics of the machine learning model;
initializing a ML scoring container in a second virtual machine instance hosted by a second computing device, wherein the ML scoring container is formed from the container image;
storing the model data in the ML scoring container;
receiving, from the user device, an execution request, wherein the execution request comprises input data;
executing second code stored in the ML scoring container using the input data to generate an output; and
transmitting the output to the user device.

10. The computer-implemented method of claim 5 , wherein the training request further comprises one or more first hyperparameter values.

11. The computer-implemented method of claim 10 , wherein causing the first virtual machine instance to execute code stored within the ML training container further comprises causing the first virtual machine instance to execute the code stored within the ML training container according to the one or more first hyperparameter values.

12. The computer-implement

Assignees

• Amazon Tech Inc

Inventors

• Faulhaber Jr Thomas Albert
• Stefani Stefano
• Thomas Owen

Classifications

• G06N20/00 Primary

  Machine learning · CPC title

• G06F9/45558

  Hypervisor-specific management and integration aspects · CPC title

• G06F9/5072 Primary

  Grid computing · CPC title

• G06N20/10

  using kernel methods, e.g. support vector machines [SVM] · CPC title

• G06F2009/4557

  Distribution of virtual machine instances; Migration and load balancing · CPC title