The field of model watermarking and fingerprinting is rapidly evolving, with a focus on developing innovative methods to verify the provenance and ownership of AI-generated content. Recent research has introduced novel watermarking techniques, such as guidance watermarking for diffusion models and ensemble frameworks for unbiased language model watermarking, which have shown promising results in terms of robustness and detectability. Additionally, there has been a growing interest in model fingerprinting, with methods like SeedPrints and LLMPrint demonstrating the ability to uniquely identify and verify the origin of large language models. Noteworthy papers in this area include 'Guidance Watermarking for Diffusion Models' and 'SeedPrints: Fingerprints Can Even Tell Which Seed Your Large Language Model Was Trained From', which have made significant contributions to the development of robust and reliable model watermarking and fingerprinting techniques.
Advances in Model Watermarking and Fingerprinting
Sources
Guidance Watermarking for Diffusion Models
Taught Well Learned Ill: Towards Distillation-conditional Backdoor Attack
An Ensemble Framework for Unbiased Language Model Watermarking
Analyzing and Evaluating Unbiased Language Model Watermark
Model Correlation Detection via Random Selection Probing
Watermarking Diffusion Language Models
Of-SemWat: High-payload text embedding for semantic watermarking of AI-generated images with arbitrary size
Fingerprinting LLMs via Prompt Injection
Reliability Crisis of Reference-free Metrics for Grammatical Error Correction
RE$^2$: Improving Chinese Grammatical Error Correction via Retrieving Appropriate Examples with Explanation
SeedPrints: Fingerprints Can Even Tell Which Seed Your Large Language Model Was Trained From
Are Robust LLM Fingerprints Adversarially Robust?
MOLM: Mixture of LoRA Markers
Fast, Secure, and High-Capacity Image Watermarking with Autoencoded Text Vectors
Detecting Post-generation Edits to Watermarked LLM Outputs via Combinatorial Watermarking
ZK-WAGON: Imperceptible Watermark for Image Generation Models using ZK-SNARKs
Knowledge Distillation Detection for Open-weights Models