Synthesize Effective Ensemble Based Dynamic Defenses to Adversarial Attacks

Description:

Reference #: 01490

The University of South Carolina is offering licensing opportunities for Synthesize Effective Ensemble Based Dynamic Defenses to Adversarial Attacks

Background: 

A lot of learning-based AI techniques are very prone to adversarial attacks. However, these attacks are detectable and there are algorithms that can defend against adversarial attacks. These existing defenses to attacks do not change at deployment time. It is possible to introduce “ensemble-based defenses”.

Invention Description:

The proposed invention is an ensemble-based defense for learning based AI. This method works by selecting a subset of weak defenses dynamically from a large array of choices. These defenses are adaptive to real-time attacks and are smaller without compromising performance.

Potential Applications:

All industries using Machine Learning can benefit from this technology, especially public technologies with need for protection such as healthcare, insurance, or self-driving cars.

Advantages and Benefits:

Smaller ensembles without compromising effectiveness of defense; uses less resources/ Ensembles that are adaptive to real-time attacks

Patent Information:
Title App Type Country Serial No. Patent No. File Date Issued Date Expire Date Patent Status
SYNTHENA Trademark Application United States 90/342,518   11/25/2020     Notice of Allowance
A System and Method for Synthesizing Dynamic Ensemble-Based Defenses to Counter Adversarial Attacks Utility United States 17,487,502   9/28/2021     Published
Category(s):
Software and Computing
For Information, Contact:
Technology Commercialization
University of South Carolina
technology@sc.edu
Inventors:
Biplav Srivastava
Jianhai Su
Ying Meng
Pooyan Jamshidi Dermani
Jason O'Kane
Keywords:
Adversarial attacks
AI safety
Ensemble based defenses
© 2024. All Rights Reserved. Powered by Inteum