Trusted Processing Device for Encrypted ML Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The increasing training cost and the need to protect sensitive data in machine learning models pose challenges in sales and deployments, necessitating secure computing to prevent secret data leakage during model usage.
Innovation Solution
A device comprising a trusted processing device with a security circuit and processing element array performs secure computing operations on encrypted machine learning models, using a first public key and private key for decryption and secure communication, along with oblivious memory controllers to protect data integrity and access patterns.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If machine learning models are provided to users for deployment, then model utility and productivity are improved, but the risk of secret data leakage increases
Solution Approach 1:
The system segments the machine learning model into encrypted portions that can be independently managed and distributed. The model is divided into multiple components that are encrypted separately, allowing selective deployment while maintaining security control over each segment.
Solution Approach 2:
A trusted execution environment acts as an intermediary between the model provider and user. This intermediate layer decrypts and executes the model in a secure sandboxed environment, preventing direct access to the model weights and structure by the user while still enabling model functionality.
2Measurement precision
If sensitive training data is used for model development, then model accuracy and performance are improved, but the risk of sensitive information exposure increases
Solution Approach 1:
The trusted execution environment serves as a mediator that processes sensitive training data without exposing it to external systems. Data can be fed into the secure environment for model training or inference, and only the results (not the raw sensitive data) are output, preventing information leakage while maintaining model accuracy.
Solution Approach 2:
The system creates an inert or isolated computing environment where sensitive data is processed in a controlled sandbox. This isolated environment prevents sensitive information from being accessed by malicious external systems while still allowing legitimate processing operations to occur.
3Reliability
If model weights and structure are protected through encryption, then security is improved, but the complexity of model deployment and operation increases
Solution Approach 1:
The system implements self-service mechanisms where the trusted execution environment automatically handles decryption and execution of encrypted model components without requiring manual intervention. The security management is automated through the secure environment's inherent capabilities, reducing the operational burden on users despite the underlying complexity.
Solution Approach 2:
The trusted execution environment provides a universal platform that can handle multiple types of encrypted model formats and security protocols through a single interface. This multi-functional approach allows the system to manage various encryption schemes and model types without proportionally increasing deployment complexity for each case.
Data Source
AI summary
A device for performing machine learning model is provided. The device comprises a computer and a trusted processing device. The computer receives a random seed generated by an electronic device. The trusted processing device comprises a security circuit and a processing element array. The security circuit is embedded with a first public key and a first private key. The security circuit generates a proof according to the random seed, the first public key and the first private key. The computer outputs the proof and the first public key to the electronic device. The electronic device determines whether to output an encrypted machine learning model to the computer. The security circuit decrypts the encrypted machine learning model to generate a machine learning model. The processing element array performs a secure computing operation of an inference of the machine learning model.