Bell Labs Internship on Decentralized, TEE-based Collaborative ML Training (PhD

at  Nokia

FAMILY DESCRIPTION

Pure Research (PR) comprises fundamental systematic research of exploratory nature for the development of new products and services as well as disruptive technologies superseding conventional approaches. In contrast to applied R&D, PR covers the provision of a comprehensive understanding of scientific insights without directed application towards products, policies, or operational processes.

SUBFAMILY DESCRIPTION

A Bell Labs Internship on VM-based TEE technology for collaborative ML training (PhD) #sandboxing #TEE #trusted-execution-environment #gpu

SHORT INTERNSHIP DESCRIPTION

Trusted execution environment (TEE) technology enables a secure enclave in a main processor. It helps code and data loaded inside the enclave to be protected with respect to confidentiality and integrity. This TEE technology has been gradually used to safeguard the collaborative ML system where data and models need to be kept confidential and integral. Such a TEE-based collaborative ML training system usually requires a centralized architecture where data and models are all submitted and run in one infrastructure with homogenous TEE configurations. In reality, however, different data owners may have different data policies. For instance, data must not leave EU due to GDPR regulations, or data must not leave a corporate network due to company policies, etc. Therefore, a new architecture for TEE-based collaborative ML training system is needed where data and models can be run on premises. This internship is to understand the current TEE-based collaborative ML systems, and design a new architecture that can be decentralized and can work with various wide-area network effects and with different heterogenous TEE configurations.

• Understand existing TEE-based collaborative ML systems.
• Design and develop a TEE-based collaborative ML system that can run in a decentralized manner with similar performance as in a centralized system.
• Report the results and findings to a top-tier academic venue.

QUALIFICATIONS

• Student enrolled in Ph.D. Computer Science/Engineering.
• Programming skills in C, Python, Pytorch/Tensorflow.
• Good system building skills.
• Language skills: English
• Strong publication record is a big plus.
  Duration : flexible, to be agreed (typically 3-4 months), starting time flexible
  Location : Stuttgart (Germany)

Please refer the Job description for details