January 01, 2024
AVATAR integrates our National Science Foundation (NSF) Small Business Innovation Research (SBIR) derived Blockchain solution (US Patent 11, 689,660 B2) to address adversarial injection in algorithms or datasets, with Microsoft’s Project Bonsai. Project Bonsai, currently in preview, works with hyper realistic simulation environments to train a Deep Reinforcement Learning (DRL) Agent, to learn to make critical decisions autonomously based on millions of iterations across custom scenarios.
By doing so, we improved the genetic diversity of the data and models to unpredictability which can harden data and ML/AI in the face of adversarial attacks.
Key Objectives
Improve the genetic diversity of the data and models to increase data and model unpredictability which can harden data and models in the face of adversarial attacks
Leverage Blockchain governance framework to address adversarial injection in algorithms and datasets, and side-channel source information about how, when, and why the AI was trained
Leverage DoD/PKI certificates to ensure non-repudiation within the ML lifecycle and attribution within the AI/ML supply chain
Detect synthetic behavior within ML/AI models and data science notebooks for suppression, and enable statistical identification of similarities in code and forensic signatures of developers that have adversarial traits
Other Relevant Experience:
© 2024 Forward Edge-AI, Inc. All rights reserved.
