Search papers, labs, and topics across Lattice.
This paper introduces FedOT, a novel framework designed for ownership verification and leakage tracing in Federated Learning (FL) of Latent Diffusion Models (LDMs). By implementing a chunked watermarking approach that separates ownership verification from client identification, and employing Latent Vector Transformation (LVT) to secure against VAE replacement attacks, FedOT effectively mitigates risks associated with unauthorized model distribution. Experimental results show that FedOT significantly outperforms existing methods in both ownership verification and traceability, ensuring the integrity of LDMs in federated settings.
FedOT not only verifies ownership but also traces model leakage back to malicious clients, a critical advancement in federated learning security.
Training Latent Diffusion Models (LDMs) within Federated Learning (FL) has attracted increasing attention due to its ability to combine the powerful generative capacity of LDMs with the privacy-preserving properties of FL. However, FL requires sharing the global model with multiple participants, which risks unauthorized model distribution or resale by malicious clients. While an intuitive approach is to adopt existing VAE-based watermarking techniques for LDMs in FL, this strategy falls short in addressing such threats due to two fundamental challenges: (1) Existing methods support ownership verification but lack the ability to trace model leakage to a specific malicious client; (2) VAE-based watermarks are vulnerable, as they can be removed simply by replacing the decoder with a clean counterpart. In this paper, we propose FedOT, the first framework for ownership verification and leakage tracing in federated LDMs. Specifically, to address the first challenge, we design a chunked watermark, where the first part is for ownership verification, and the second part is used for client identification. Furthermore, to overcome the second challenge and secure the model against VAE replacement attack, we introduce Latent Vector Transformation (LVT), which strengthens the connection between the VAE and U-Net latent spaces by modifying the original latent distribution of the VAE. Consequently, any attempt to replace the VAE for watermark removal leads to significant image quality degradation, making the LDM model unusable. Extensive experiments demonstrate that FedOT achieves superior performance in both ownership verification and traceability. Project page: https://spyzixuan.github.io/FedOT/.
NUS