Abstract
Machine learning (ML) models for various purposes may be made available via cloud-based execution platforms. The data owner, the model provider, and the execution platform where a model is deployed may be different entities. In this configuration, it is necessary to establish trust between the model provider, the execution platform, and the user to enable secure and private use of machine learning models. This disclosure describes an execution platform that can be verified and authenticated by a data owner as a model provider using techniques such as confidential computing and/or trusted execution environment. The execution platform enables the model provider to make their models available for use by data owners. Data owners can provide encrypted data to the execution platform and specify a ML model to be applied to the data. The execution platform implements an instance of the ML model and enables secure access to the user data without the model provider being able to access the data. Thus, users can obtain the benefit of using a ML model of their choice via the execution platform and model providers can provide models in a secure marketplace, while preserving the privacy and security for both entities.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Glisse, Jérôme, "Secure and Private Access to Machine Learning Models via Cloud-based Execution Platform", Technical Disclosure Commons, (April 26, 2023)
https://www.tdcommons.org/dpubs_series/5845