Controlling execution of machine learning models

Patent metadata

Field	Value
Publication number	US-12470381-B2
Application number	US-202018249164-A
Country	US
Kind code	B2
Filing date	Oct 29, 2020
Priority date	Oct 29, 2020
Publication date	Nov 11, 2025
Grant date	Nov 11, 2025

Abstract

In an example, an apparatus is described. The apparatus comprises processing circuitry comprising a control module. The control module determines whether a computing device communicatively coupled to the control module is in a specified state for executing a machine learning model controlled by a third party entity. In response to determining that the computing device is in the specified state, the control module is to send, to an attestation module in a data processing pipeline associated with the computing device, an indication that the computing device is in the specified state.

First claim

The invention claimed is:

1 . An apparatus comprising processing circuitry, the processing circuitry comprising:
an attestation module to:
operate in a control plane of a data processing pipeline, and
generate a cryptographically signed attestation based on a model execution specification that defines a specified state for executing a machine learning model controlled by a third party entity; and
a control module to:
operate in the control plane of the data processing pipeline,
determine whether a computing device communicatively coupled to the control module is in the specified state, and
send, to the attestation module in response to determining that the computing device is in the specified state, an indication that the computing device is in the specified state.

2 . The apparatus of claim 1 , wherein, in response to determining that the computing device is not in the specified state, the control module is to send an instruction to the computing device to set-up the computing device in accordance with the specified state.

3 . The apparatus of claim 1 , wherein the attestation module is to attest to the computing device being in the specified state by sending a signed statement comprising the indication to the third party entity.

4 . The apparatus of claim 1 , wherein the control module is to verify a signature applied to the machine learning model and/or an associated model specification by the third party entity against a public key associated with the third party entity, and where the control module is to provide verification that the signature matches the public key via the indication.

5 . The apparatus of claim 4 , wherein, in response to determining that the computing device is in the specified state and that either the machine learning model or the model execution specification or both is verified against the public key associated with the third party entity, the control module is to indicate, via the indication, that the machine learning model is executable by the computing device.

6 . The apparatus of claim 1 , wherein the indication sent to the attestation module includes a timestamp corresponding to when the computing device was determined to be in the specified state.

7 . The apparatus of claim 1 , wherein the indication comprises a hash of the machine learning model or the model execution specification to be executed by the computing device.

8 . The apparatus of claim 1 , wherein the control module is to determine whether the computing device supports an incremental learning procedure prior to sending the indication to the attestation module.

9 . The apparatus of claim 1 , wherein the control module is to determine whether the computing device supports a distributed learning procedure prior to sending the indication to the attestation module.

10 . The apparatus of claim 1 , wherein the control module is to initiate execution of a test policy through the machine learning model to validate setup of the computing device prior to sending the indication to the attestation module.

11 . A tangible machine readable medium comprising instructions which, when executed by at least one processor, cause the processor to:
verify, using a cryptographic key, a digital signature associated with a model execution specification that defines one or more conditions for executing a machine learning model controlled by a third party entity; determine whether a computing device under control of the processor is capable of operating in accordance with the model execution specification; and in response to determining that the computing device is capable of operating in accordance with the model execution specification, cause the computing device to establish a data processing pipeline for executing the machine learning model in accordance with the model execution specification.

12 . The tangible machine readable medium of claim 11 , wherein the instructions are to cause the processor to:
obtain information from a data handling module of the computing device in the data processing pipeline regarding a state of the data handling module; and determine whether or not the data handling module is capable of operating in accordance with the model execution specification.

13 . The tangible machine readable medium of claim 11 , wherein the instructions are to cause the computing device to load the machine learning model by setting up a communication channel between a memory storing information regarding the machine learning model and the computing device, and where the instructions are to cause transfer of the information to the computing device.

14 . The tangible machine readable medium of claim 13 , wherein the instructions are to cause the computing device to execute a test policy through the machine learning model at load time to determine whether or not the computing device is set up in accordance with the model execution specification.

15 . A method, comprising:
generating, by a third party entity, a model execution specification comprising one or more cryptographically verifiable conditions for executing a machine learning model controlled by a third party entity; cryptographically signing, using a private key associated with the third party entity, a package of information comprising the model execution specification and the machine learning model; and sending the signed package to a control module in control of a computing device for executing the machine learning model, wherein the control module, upon verifying a cryptographic signature associated with the signed package, determines whether the computing device satisfies the one or more cryptographically verifiable conditions.

16 . The method of claim 15 , comprising encrypting the information regarding the machine learning model under a public key of the control module prior to sending the information to the control module.

17 . The method of claim 15 , comprising, in response to receiving an attestation that the computing device complies with the one or more cryptographically verifiable conditions, causing the control module to facilitate execution of the machine learning model by the computing device in accordance with the one or more cryptographically verifiable conditions.

18 . The method of claim 17 , comprising, in response to the attestation comprising an indication that the machine learning model was verified against a public key associated with the private key, verifying whether or not the control module has set up the computing device in accordance with the one or more cryptographically verifiable conditions.

19 . The method of claim 15 , wherein the model execution specification a condition comprises from the group consisting of:
a description of a specified data processing pipeline for executing the machine learning model; an associated hash of the machine learning model; a test procedure indicator; and an incremental learning procedure indicator and/or a distributed learning procedure indicator.

20 . The method of claim 15 , wherein the information and/or the one or more cryptographically verifiable conditions comprises a public key of a service provider authorized by the third party entity to receive an encrypted result obtained by executing the machine learning model on the computing device.

21 . The method of claim 15 , wherein the information comprises a public key of a service provider authorized by the third party entity to receive an encrypted result obtained by executing the

Assignees

• Hewlett Packard Development Co

Inventors

• Baldwin Adrian John
• Dalton Christopher Ian
• Belgarric Pierre
• Plaquin David
• Ellam Daniel Cameron

Classifications

• G06N20/00 Primary

  Machine learning · CPC title

• H04L9/50

  using hash chains, e.g. blockchains or hash trees · CPC title

• H04L9/0897

  involving additional devices, e.g. trusted platform module [TPM], smartcard or USB · CPC title

• H04L9/3247

  involving digital signatures · CPC title

• G06F11/28

  by checking the correct order of processing (G06F11/08 - G06F11/26 take precedence; monitoring patterns of pulse trains H03K5/19) · CPC title