Machine learning platform with model storage

Patent metadata

Field	Value
Publication number	US-11836268-B2
Application number	US-202017062409-A
Country	US
Kind code	B2
Filing date	Oct 2, 2020
Priority date	Oct 2, 2020
Publication date	Dec 5, 2023
Grant date	Dec 5, 2023

Abstract

A request to perform a prediction using a machine learning model of a specific entity is received. A specific security key for the machine learning model of the specific entity is received. At least a portion of the machine learning model is obtained from a multi-tenant machine learning model storage. The machine learning model is unlocked using the specific security key and the requested prediction is performed. A result of the prediction is provided from a prediction server.

First claim

What is claimed is:

1. A method, comprising:
receiving a request to perform a prediction using a machine learning model of a specific entity;
receiving a specific security key for the machine learning model of the specific entity;
obtaining at least a portion of the machine learning model from a multi-tenant machine learning model storage, including by storing a utilization status of the machine learning model, wherein the utilization status associates an identifier of a prediction server utilizing the machine learning model;
unlocking the machine learning model using the specific security key;
performing the requested prediction; and
providing a result of the prediction from the prediction server.

2. The method of claim 1 , wherein the specific security key for the machine learning model of the specific entity is stored in a single-tenant storage.

3. The method of claim 1 , wherein the requested prediction utilizes the machine learning model and an input data set, and wherein the input data set is retrieved from a single-tenant storage.

4. The method of claim 1 , wherein the multi-tenant machine learning model storage is utilized to:
receive an incremental update to the machine learning model; and
apply the incremental update to the machine learning model.

5. The method of claim 4 , wherein the multi-tenant machine learning model storage is utilized to:
unlock the machine learning model using a first security key prior to applying the incremental update; and
lock the updated machine learning model using a second security key after applying the incremental update.

6. The method of claim 4 , wherein the multi-tenant machine learning model storage is utilized to store a version information associated with the incremental update to the machine learning model.

7. The method of claim 1 , wherein the multi-tenant machine learning model storage is communicatively connected to the prediction server via a network attached storage, a storage area network, or a mounted storage device.

8. The method of claim 1 , further comprising:
purging data of a first tenant from the prediction server, wherein the first tenant is different from the specific entity.

9. The method of claim 8 , wherein the data of the first tenant includes a security key of the first tenant, a portion of a machine learning model of the first tenant, an input data of the first tenant, an intermediate prediction result of the first tenant, or a prediction result of the first tenant.

10. The method of claim 1 , wherein the prediction server is included in a cluster of prediction servers.

11. The method of claim 1 , wherein the utilization status indicates that that the machine learning model has been checked out by the prediction server.

12. The method of claim 1 , wherein the utilization status includes a version identifier of the machine learning model.

13. The method of claim 1 , wherein the result of the prediction is encrypted using an encryption key of the specific entity.

14. The method of claim 1 , wherein the prediction server is configured to process requests to perform predictions from multiple tenants one at a time in a sequential order.

15. The method of claim 1 , wherein a decrypted portion of the machine learning model is cached in a transitory memory of the prediction server.

16. The method of claim 1 , wherein the specific security key is a private encryption key of a private-public key pair.

17. A method, comprising:
determining to perform a prediction using a machine learning model of a specific entity;
obtaining a specific security key for the machine learning model of the specific entity;
providing to a prediction server, the specific security key and a request to perform the prediction, wherein the prediction server is configured to obtain at least a portion of the machine learning model from a multi-tenant machine learning model storage and the prediction server is configured to utilize the specific security key to unlock the machine learning model, and wherein a stored utilization status associates an identifier of the prediction server utilizing the machine learning model; and
receiving a result of the prediction from the prediction server.

18. The method of claim 17 , wherein the specific security key for the machine learning model of the specific entity is obtained from a storage dedicated for the specific entity.

19. A method, comprising:
receiving a request to train a machine learning model of a specific entity;
receiving a specific security key of the specific entity;
obtaining at least a portion of training data from a single-tenant data storage;
training the machine learning model using at least the obtained portion of the training data;
encrypting the machine learning model using the received specific security key of the specific entity;
providing the encrypted machine learning model for storage in a multi-tenant machine learning model storage; and
obtaining at least a portion of the machine learning model from the multi-tenant machine learning model storage, including by storing a utilization status of the machine learning model, wherein the utilization status associates an identifier of a prediction server utilizing the machine learning model.

20. The method of claim 19 , further comprising:
purging a local copy of the at least portion of the training data obtained from the single-tenant data storage; and
purging a local copy of the received specific security key of the specific entity.

Assignees

• Servicenow Inc

Inventors

• Mehta Virendra Kumar
• Palapudi Sriram

Classifications

• G06F21/6227 Primary

  where protection concerns the structure of data, e.g. records, types, queries · CPC title

• G06F8/658

  Incremental updates; Differential updates · CPC title

• G06F18/214

  Generating training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title

• G06F21/602

  Providing cryptographic facilities or services · CPC title

• G06N20/00

  Machine learning · CPC title