Title:
WATERMARK INFORMATION EMBEDDING METHOD AND APPARATUS
Document Type and Number:
WIPO Patent Application WO/2021/083111
Kind Code:
A1
Abstract:
A watermark information embedding method and apparatus. The method comprises: training an embedding neural network model by means of weight information of a target neural network model and target watermark information to be embedded to the target neural network model, updating the weight information of the target neural network model according to target watermark embedding data provided by the embedding neural network model, and obtaining a target neural network model embedded with the target watermark information. Because an embedding neural network comprises a plurality of neural network layers, the method increases the complexity of a watermark embedding process, and avoids the problem of poor robustness of watermark information of existing neural network models against watermark attack means such as overwrite attack and model compression.
Inventors:
WANG TIANHAO (CN)
LIU YONGLIANG (CN)
ZHANG QI (CN)
LIU YONGLIANG (CN)
ZHANG QI (CN)
Application Number:
PCT/CN2020/123888
Publication Date:
May 06, 2021
Filing Date:
October 27, 2020
Export Citation:
Assignee:
ALIBABA GROUP HOLDING LTD (CN)
International Classes:
G06F21/16; G06N3/08
Domestic Patent References:
| WO2018180263A1 | 2018-10-04 |
Foreign References:
| CN110168576A | 2019-08-23 | |||
| CN110276708A | 2019-09-24 | |||
| CN101389027A | 2009-03-18 | |||
| JP2019053541A | 2019-04-04 | |||
| CN109919303A | 2019-06-21 |
Other References:
See also references of EP 4053718A4
Attorney, Agent or Firm:
BEIJING SANYOU INTELLECTUAL PROPERTY AGENCY LTD. (CN)
Download PDF: