LVQMark: Robust Time-Series Watermarking via Local Tokenization

LVQMark addresses watermarking instability in multivariate time series caused by post-editing attacks. By using a locally tokenized L-VQVAE generative model, the system ensures each watermarked unit depends only on a bounded temporal neighborhood, preventing null distribution drift.