AI Model Activation Function Periodic Update for Adversarial Security

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for securing artificial intelligence models against adversarial attacks, such as encrypting entire or partial layers, face challenges like increased data storage requirements and decreased processing rates, which are inefficient and power-consuming.

Innovation Solution

An electronic apparatus that updates the activation function of an AI model by adding periodic functions at different time intervals, allowing the model to adapt without encryption, thereby enhancing security without the need for extensive data storage or increased power consumption.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If encryption is applied to the artificial intelligence model or its layers, then security is improved, but data storage space increases and processing rate decreases

Engineering Contradiction:
ImprovesecurityVSAvoidprocessing rate
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The patent changes the parameter of the activation function by adding a periodic function with time-varying characteristics. This modifies the activation function's behavior dynamically over time without requiring encryption, thus maintaining processing speed while improving security against adversarial attacks that rely on static model analysis

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The activation function is transformed from a static function to a dynamic one by incorporating a periodic function that changes over time. This dynamic characteristic makes the model harder to analyze and attack while avoiding the overhead of encryption operations, thereby maintaining high processing rates

Inventive Principle:
Principle #15Dynamics

2Reliability

If encryption is applied to the artificial intelligence model or its layers, then security is improved, but power consumption increases

Engineering Contradiction:
ImprovesecurityVSAvoidpower consumption
Core Design Contradiction:
ReliabilityVSUse of energy by moving object

Solution Approach 1:

Instead of using computationally intensive encryption operations, the patent changes the activation function's parameters by adding a periodic function. This approach achieves security improvement through mathematical transformation rather than cryptographic operations, significantly reducing power consumption while maintaining model functionality

Inventive Principle:
Principle #35Parameter changes

3Reliability

If the activation function is updated periodically, then security against adversarial attacks is improved, but model training complexity increases

Engineering Contradiction:
ImprovesecurityVSAvoidmodel training complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The periodic function is designed and added in advance to the activation function before deployment. This preliminary action allows the security mechanism to be built-in without requiring complex real-time adjustments or retraining, simplifying the overall training process while maintaining security benefits

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentUS20230409701A1Electronic apparatus and control method thereof
Publication Date: 2023.12.21 SAMSUNG ELECTRONICS CO LTD
  • US20230409701A1 patent drawing
  • US20230409701A1 patent drawing
  • US20230409701A1 patent drawing

AI summary

An electronic apparatus includes a memory storing an artificial intelligence model and at least one processor configured to identify a first activation function used in at least one layer of the artificial intelligence model, obtain a second activation function by adding a first periodic function corresponding to a first time interval to the first activation function, apply the second activation function to an output layer during the first time interval, obtain a third activation function by adding a second periodic function corresponding to a second time interval to the first activation function, and update the artificial intelligence model by applying the third activation function to the output layer during the second time interval.