Student Publications [Scholarly]

A Survey of Vulnerabilities and Emerging Defenses in GANs, VAEs, and Transformers

Document Type

Conference Proceeding

Abstract

Generative Artificial Intelligence is the foundation of an increasingly diverse array of applications - text and image synthesis, multimodal reasoning, workflow of autonomous agents. Innovation is being taken on by recently published models like Transformers, GANs, etc. But even now that these models are realizing unparalleled ability, they are revealing a more complicated vulnerability terrain. It works on adversarial training, alignment tuning, latent regularization, model firewalls, ensemble smoothing, and provenance verification and identifies tradeoffs connected with these methods. Although no experimental data has been reported in this article, the conceptual frameworks of generative AI robustness help in the future of research on multimodal robustness and generative model defensive design, as well as supply chain verification. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2027.

Publication Title

Communications in Computer and Information Science

Publication Date

2027

Volume

2938 CCIS

First Page

511

Last Page

518

ISBN

9783032221957

DOI

10.1007/978-3-032-22196-4_38

Keywords

adversarial attacks, Generative Artificial Intelligence, large language models, transformers

Link to Full Text

Share

COinS