This paper explores the role of artificial intelligence (AI) in the metaverse, particularly focusing on collaborative deep learning (CDL). It identifies potential security risks, such as malicious participants and GAN-attacks, that could compromise the safety of the CDL process or the data involved. To address these issues, the authors propose an adversary detection-deactivation method aimed at limiting and isolating potential malicious participants and preventing harmful access. The method has been tested on a Multiview CDL case, and the results demonstrate its effectiveness in protecting the model and data.
Publication date: 5 Jan 2024
Project Page: Not provided
Paper: https://arxiv.org/pdf/2401.01895
