Obfuscation Unit for Secure AI Data Transmission

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Data transmission for AI training data and models is not adequately protected, and existing cryptographic key-based solutions are slow and impractical, necessitating a system to obscure data transmissions for data processing accelerators without relying on cryptography.

Innovation Solution

A method and system using an obfuscation unit in data processing accelerators to obfuscate and de-obfuscate AI models and training input data, enabling secure communication between hosts and data processing accelerators through an obfuscation kernel algorithm, providing multiple layers of protection without the need for cryptographic key-based solutions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If cryptographic key-based solutions are used to protect data transmission, then data security is improved, but processing speed deteriorates and the system becomes impractical

Engineering Contradiction:
Improvedata securityVSAvoidprocessing speed
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The patent replaces cryptographic key-based security mechanisms with a hardware-based obfuscation system using an obfuscation unit that applies obfuscation kernels to data transmissions. This substitution eliminates the computational overhead of cryptographic operations while maintaining security through hardware-enforced obfuscation and de-obfuscation processes.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent transforms security from a cryptographic parameter space to an obfuscation parameter space by using obfuscation kernels with specific parameters (alpha, beta, gamma) that control the obfuscation process. This allows fast mathematical operations to provide security without the slowness of traditional cryptography.

Inventive Principle:
Principle #35Parameter changes

2Productivity

If data is transmitted without protection for DP accelerators, then processing speed is maintained, but data security deteriorates and data may be leaked to untrusted parties

Engineering Contradiction:
Improveprocessing speedVSAvoiddata security
Core Design Contradiction:
ProductivityVSReliability

Solution Approach 1:

The patent introduces an obfuscation unit as an intermediary component between the host and DP accelerator. This unit applies obfuscation to data transmissions without interfering with the underlying fast processing capabilities of the DP accelerator, thus maintaining productivity while adding security protection.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Reliability

If an obfuscation unit is added to the DP accelerator, then data security is improved, but device complexity increases

Engineering Contradiction:
Improvedata securityVSAvoidsystem complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent merges the obfuscation unit with the DP accelerator into a single integrated device. The obfuscation unit shares hardware resources and communication interfaces with the DP accelerator, reducing overall system complexity compared to having separate security and processing systems.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The obfuscation unit is designed to handle multiple security functions including obfuscation, de-obfuscation, and security management for various data types (training data, models, inference outputs) within the DP accelerator ecosystem, reducing the need for multiple specialized security components.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Data Source

PatentEP3793162B1Data transmission with obfuscation using an obfuscation unit for a data processing (DP) accelerator
Publication Date: 2023.05.10 BAIDU USA LLC
  • EP3793162B1 patent drawingFigure 1
  • EP3793162B1 patent drawingFigure 2
  • EP3793162B1 patent drawingFigure 3

AI summary

According to one embodiment, a host communicates with a data processing (DP) accelerator using an obfuscation scheme. The DP accelerator receives a training request from a host, the training request includes an obfuscated data that includes one or more AI models and/or training input data. The DP accelerator de-obfuscates, by an obfuscation unit of the DP accelerator, the obfuscated data to obtain the one or more AI models. The DP accelerator trains the one or more AI models based on the training input data.