Securing Artificial Intelligence (SAI); Collaborative Artificial Intelligence

Abstract

This work item covers security aspects of pervasive and collaborative artificial intelligence (AI) (e.g., federated learning, transfer learning, distributed reinforcement learning, decentralized machine learning), where multiple AI agents are pervasively distributed in different places but interact with each other to work on the same or different AI tasks. These AI agents exchange AI models and/or inferred knowledge. The work item investigates use cases of security for pervasive and collaborative AI, and analyzes potential security concerns such as trust and communications among those AI agents. Then the work item provides technical recommendations on approaches to mitigate these issues including their limitations.