Secure Learning Model Execution with Pre-Decryption Falsification Detection

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing methods for maintaining confidentiality of learning model information do not effectively detect falsification, leaving models vulnerable to unauthorized modifications.

Innovation Solution

An information processing device with a rich execution unit and a secure execution unit, equipped with decryption, operation execution, falsification detection, and output control units, ensures confidentiality and detects falsification by decrypting encrypted model information and verifying its authenticity before executing operations.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If encrypted learning model information is decrypted and executed in a secure execution environment, then confidentiality of model information is maintained, but falsification detection capability is insufficient

Engineering Contradiction:

Improveconfidentiality of model informationVSAvoidfalsification detection capability

Core Design Contradiction:

ReliabilityVSDifficulty of detecting and measuring

Solution Approach 1:

The patent applies preliminary action by performing falsification detection on the encrypted model information before decryption and execution. The falsification detection unit checks whether the encrypted model information is falsified information in advance, preventing potentially malicious or corrupted models from being deployed. This resolves the contradiction by adding a detection mechanism that operates prior to the confidentiality-protecting decryption process.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent introduces an intermediary falsification detection unit that acts as a mediator between the encrypted model information and the secure execution environment. This intermediary component verifies the authenticity of the model information without compromising the confidentiality maintained by the secure execution environment, thus enabling both confidentiality and falsification detection simultaneously.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Reliability

If a secure execution environment is used to decrypt and execute model information, then model confidentiality is protected, but the system complexity increases

Engineering Contradiction:

Improveconfidentiality protectionVSAvoidsystem complexity

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The patent merges the falsification detection function with the secure execution environment, combining multiple security functions (confidentiality protection and falsification detection) into a unified system. The falsification detection unit operates within or in conjunction with the secure execution environment, avoiding the need for separate independent systems and thus reducing overall complexity while maintaining both security functions.

Inventive Principle:

Principle #5Merging (Combining)

Solution Approach 2:

The secure execution environment is designed to perform multiple functions: it provides confidentiality protection through encryption, executes model operations securely, and incorporates falsification detection capabilities. This multi-functionality reduces the need for separate dedicated systems for each function, thereby managing system complexity while achieving comprehensive security.

Inventive Principle:

Principle #6Universality (Multi-functionality)

3Difficulty of detecting and measuring

If falsification detection is added to the secure execution unit, then detection capability is improved, but processing time increases

Engineering Contradiction:

Improvefalsification detection capabilityVSAvoidprocessing time

Core Design Contradiction:

Difficulty of detecting and measuringVSLoss of time

Solution Approach 1:

The falsification detection is performed as a preliminary action on the encrypted model information before decryption and execution. By conducting the detection in advance, the system avoids time-consuming operations during the actual model execution phase. The detection unit efficiently verifies authenticity using cryptographic operations that complete quickly, minimizing the time added to the overall processing pipeline.

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS12355883B2Information processing device, information processing method, and computer readable medium

Publication Date: 2025.07.08 MITSUBISHI ELECTRIC CORP

AI summary

A rich execution unit (110) executes an application (111) while executing a rich OS. A secure execution unit (120) executes a target operation, which is an operation based on a learning model, while executing a secure OS. A decryption processing unit (122) decrypts model information from encrypted model information of the learning model when execution of the target operation is requested by the application. An operation execution unit (121) executes the target operation using the model information. A falsification detection unit (123) determines whether the encrypted model information is falsified information. An output control unit (124) passes an execution result of the target operation to the application when it is determined that the encrypted model information is not the falsified information.