SDS 843: Safe, Fast and Efficient AI — with Protopia’s Dr. Eiman Ebrahimi

Proactive Security

Protopia AI's proactive approach to data security focuses on minimizing the utility of compromised data. Eiman Ebrahimi explains that their Stained Glass Transform solution transforms raw data into a format that is difficult for unauthorized users to exploit, even if accessed 1. This method leverages the expansive representational spaces of machine learning models to create a moving target for potential attackers. Ebrahimi highlights the importance of securing data at the platform level, as even trusted models can be compromised due to human errors or system vulnerabilities 2.

The idea is one of utilizing essentially the fundamentals of machine learning models and the fact that machine learning models live in fairly large, often representational, spaces.

--- Eiman Ebrahimi

This proactive stance is crucial in an era where data breaches are increasingly common.

   

Multitenancy Challenges

Multitenancy in AI systems presents unique data security challenges, as multiple users share the same hardware infrastructure. Eiman Ebrahimi points out that this shared environment requires a collective responsibility for security, as one user's oversight can compromise the entire system 3. The concept of multitenancy involves different users sending requests to the same hardware, which can occur at various levels, from chip to rack 4.

Multitenancy is a way that we've always looked at making the use of systems more efficient for inferencing of machine learning.

--- Eiman Ebrahimi

This efficiency, however, must be balanced with robust security measures to protect sensitive data.