US 11,893,094 B1 · DCT-based watermarking scheme for deep neural networks

Abstract

The invention discloses a DNN watermarking method, comprising embedding part of the digital watermark in selected redundant elements of a deep neural network (DNN) model—without compromising the performance of the DNN. The proposed method aims for a robust watermark scheme by embedding a large watermark that can span the whole DNN model. If an adversary attempts to destroy the watermark, the whole DNN model will be destroyed. However, maximizing the hiding capacity can lead to degradation in the performance of the DNN model. In this work, this capacity-performance trade-off problem is solved using the Discrete Cosine Transform (DCT). Moreover, the DCT can work more efficiently with highly correlated data. Therefore, this work suggests segmenting the weights of the DNN model into correlated segments to fully exploit the advantages of the DCT.

Claims5

Dependent claims

A method of data hiding based on exploiting redundant information of a deep machine learning model, the method comprising:

extracting weights of the deep machine learning model;
creating homogeneous segments out of the extracted weights;
transforming the extracted weights into a frequency domain;
identifying insignificant coefficients of the frequency domain and replacing the insignificant coefficients with a watermark signal;
and transforming the modified weights back to a native domain,
adopting a discrete cosine transform (DCT) to express the deep neural network (DNN) weights vector within significant DCT coefficients, leaving a plurality of the insignificant coefficients to host part of a digital watermark,
strengthening a compaction property of the DCT when applied to correlated segments of data, and therefore, weights are segmented into correlated segments before applying the DCT.

4 more elements

Dependencies are inferred from the claim text.

DCT-based watermarking scheme for deep neural networks

Life of this patent

DCT-based watermarking scheme for deep neural networks

Life of this patent

Abstract

Claims5