Mehrdad Saberi

CS Ph.D. student @UofMaryland

[Mehrdad Saberi @MerhdadS]

@tahir_7710 @FeiziSoheil Coding theory is a diverse field, but if you're interested in error correction codes, I suggest studying popular techniques such as LDPC, Polar, Turbo, and BCH codes.

[Mehrdad Saberi @MerhdadS]

Read more about our approach and results here:
https://t.co/yY0NqE0YRi

Our code is available at:
https://t.co/3bhCvPanqd

[Mehrdad Saberi @MerhdadS]

DREW offers a vast range of applications including the detection of AI-generated content and deepfake, protecting ownership of personal media or legal documents, and digital forensics. This can mitigate some of the risks associated with the fast growth of generative AI. 2/

[Mehrdad Saberi @MerhdadS]

In our previous work (https://t.co/r2xWlb86Xa), we exposed the unreliability of watermarking techniques. However, DREW proposes a robust application of watermarks by using a large number of error control bits, ensuring robustness regardless of the severity of attacks. 3/

[Mehrdad Saberi @MerhdadS]

Wonder how to securely🔐trace the source or claim ownership of media (personal, AI-generated, art)? Traditional methods often fail against malicious alterations. Meet DREW: a robust data provenance method combining watermarking and cloud-based retrieval.🧵1/

[Soheil Feizi @FeiziSoheil]

AI giants like Google use image watermarking to combat deepfakes

We now show that:

(i) existing methods are quite vulnerable against diffusion, adversarial & spoofing attacks

(ii) imperceptible watermarks can never become reliable

Paper: https://t.co/MUgRCXtRB8

A 🧵