Neural Network Controller Precision Reduction for Brute Force Protection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Neural networks are vulnerable to unauthorized replication through brute force attacks, as existing security measures like secure elements and homomorphic encryption are either ineffective or degrade performance, and watermarking does not prevent copying.

Innovation Solution

Implementing a neural network controller that reduces the precision of output from the activation function, making it less susceptible to brute force attacks by converting high-precision outputs to lower precision, thereby obscuring changes in node weights and biases, and using a plurality of secured neural network controllers for processing to enhance security and performance.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If high-precision output is used from the activation function, then the accuracy and quality of neural network results are improved, but the neural network becomes more vulnerable to brute force attacks and parameter theft

Engineering Contradiction:
Improveoutput precisionVSAvoidvulnerability to brute force attacks
Core Design Contradiction:
Measurement precisionVSObject-affected harmful factors

Solution Approach 1:

The patent extracts only the necessary precision level from the high-precision output and removes the excessive precision information that could be exploited for brute force attacks. By outputting only the most significant bits or rounded values, the system retains functional accuracy while eliminating the harmful information that would enable parameter reverse-engineering.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent applies different precision levels to different parts of the output. Critical output values maintain high precision for functional accuracy, while less critical values are output with reduced precision to prevent attack vectors. This localized precision management allows the system to maintain overall accuracy while securing against brute force attacks on specific parameters.

Inventive Principle:
Principle #3Local quality

2Reliability

If security measures like homomorphic encryption are implemented, then the security of the neural network is improved, but the computational performance and speed are significantly degraded

Engineering Contradiction:
ImprovesecurityVSAvoidcomputational performance
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The patent uses inexpensive, computationally lightweight obfuscation techniques instead of heavy cryptographic protocols. By employing simple rounding, bit truncation, or modular arithmetic operations that are computationally cheap, the system achieves security without the performance penalty of expensive encryption schemes like homomorphic encryption.

Inventive Principle:
Principle #27Cheap short-living objects (Disposable)

Solution Approach 2:

The patent changes the precision parameter of the output values to provide security. By adjusting the number of significant bits or the rounding level dynamically, the system can control the balance between security and performance without invoking complex cryptographic transformations that would degrade computational speed.

Inventive Principle:
Principle #35Parameter changes

3Loss of information

If watermarking is used to protect the neural network, then the ability to detect copying is improved, but the prevention of copying is not achieved

Engineering Contradiction:
Improvedetection of copyingVSAvoidprevention of copying
Core Design Contradiction:
Loss of informationVSReliability

Solution Approach 1:

The patent applies preliminary anti-action by preventing the copying process itself through precision reduction, rather than merely detecting it afterward. By making the parameters unrecoverable through reduced precision output, the system proactively prevents successful copying attempts before they can be detected, thereby achieving both detection and prevention capabilities.

Inventive Principle:
Principle #9Preliminary anti-action

Data Source

PatentEP3910555B1Neural network security
Publication Date: 2025.03.26 INFINEON TECHNOLOGIES AG
  • EP3910555B1 patent drawingFigure 1
  • EP3910555B1 patent drawingFigure 2
  • EP3910555B1 patent drawingFigure 3~4

AI summary

Herein is disclosed a neural network controller, configured to implement a neural network, the neural network including: a first layer; one or more second layers; and a third layer; wherein each layer of the first layer, the one or more second layers, and the third layer includes one or more nodes; wherein at least one node of the one or more second layers is configured to provide an output value at a first level of precision; wherein the neural network controller is configured to implement a precision reduction function to reduce an output value of at least one node of the third layer to a second level of precision; and wherein the second level of precision is less precise than the first level of precision.