Inspection of algorithm containers for Sagemaker

0

We plan to provide a machine learning algorithm via a container image and are concerned about. Is it possible that other parties download the docker image for local inspection?

  • Can you clarify your question? Would you want members in your team to be able to download a custom image you have created? I'm assuming this question is specific to model monitoring.

  • Thanks Durga_S, to clarify: is it possible for unrelated parties to get the source contained in the container image? That is, if the container contains for example Python code that is not open source, would it be possible for unrelated entities to obtain the Python sources?

Norbert
asked 2 years ago306 views
2 Answers
0

As long as you have the image pushed to an ECR repository you should be fine, you will want to check access and control permissions to your repository and if you want it in a VPC or not.

AWS
answered 2 years ago
  • Yes, we are pushing to an ECR repository. One more thing I would like to clarify: We want third parties to use the container in sagemaker, but not inspect the containers. Does this align with your understanding? Thanks a lot!

0

That is, if the container contains for example Python code that is not open source, would it be possible for unrelated entities to obtain the Python sources?

Third parties who can access or download the image will be able to access the internals, i.e. via docker run -it your_secret_image /bin/bash. If you are distributing your container for use, you are distributing the contents of the container for access as well.

See also a similar question asked and answered elsewhere.

AWS
answered 2 years ago
  • Thanks, and yes, I am aware of the fact that if the image is generally available for download, then it can be inspected.

    The question is whether sagemaker algorithm containers can be used within sagemaker (that is on the AWS cloud servers) AND at the same time set to NOT be downloadable to unrelated computers.

    My idea of how it "should" work is that Sagemaker algorithm containers can be deployed to cloud nodes via the Python API, but not pulled via docker pull etc. That would protect the content of the image and still make the functionality available in sagemaker.

You are not logged in. Log in to post an answer.

A good answer clearly answers the question and provides constructive feedback and encourages professional growth in the question asker.

Guidelines for Answering Questions