Cosine Model Watermarking Against Ensemble Distillation

TL;DR

This paper introduces CosWM, a novel watermarking method designed to robustly protect models against ensemble distillation attacks, with strong theoretical guarantees and superior experimental performance.

Contribution

The paper presents CosWM, a new watermarking technique that effectively defends against ensemble distillation, outperforming existing methods with theoretical support.

Findings

CosWM achieves high watermark robustness against ensemble distillation.

Extensive experiments show CosWM outperforms state-of-the-art baselines.

CosWM provides theoretical guarantees of watermark security.

Abstract

Many model watermarking methods have been developed to prevent valuable deployed commercial models from being stealthily stolen by model distillations. However, watermarks produced by most existing model watermarking methods can be easily evaded by ensemble distillation, because averaging the outputs of multiple ensembled models can significantly reduce or even erase the watermarks. In this paper, we focus on tackling the challenging task of defending against ensemble distillation. We propose a novel watermarking technique named CosWM to achieve outstanding model watermarking performance against ensemble distillation. CosWM is not only elegant in design, but also comes with desirable theoretical guarantees. Our extensive experiments on public data sets demonstrate the excellent performance of CosWM and its advantages over the state-of-the-art baselines.