This is the public page of the Neural Network Watermarking (MPAI-NNW) standard. See the MPAI-NNW homepage.

Watermarking is a technology inherited from the multimedia domain, regrouping a family of methodological and applicative tools allowing to imperceptibly and persistently insert some metadata into an original content.

YouTube video Non-YouTube video Ppt presentation:  MPAI-NNW DV3.0

Watermarking is inherited from the multimedia domain and offers a family of methodological and application tools to imperceptibly and persistently insert some metadata into an original content.

The purpose of the MPAI Neural Network Watermarking (NNW) standard is to specify tools for qualifying an NN watermarking technologies. Neural Network Watermarking (NNW) is a Technical Specification providing the means to measure, for a given size of the watermarking payload:

  1. The ability of a neural network watermark injector to inject a payload without deteriorating the performance of the Neural Network. To this end:
    1. A pair of training and testing datasets is defined.
    2. The watermark injector is used on unwatermarked NN trained on the task.
    3. Both unwatermarked and watermarked NN are evaluated on the same testing dataset.
  2. The ability of a neural network watermarking detector/decoder to detect/decode a payload when the watermarked neural network has been modified. To this end:
    1. A pair of training and testing datasets is defined.
    2. The watermark injector is used on unwatermarked NN trained on the task.
    3. The robustness of detector/decoder is assess by evaluating their ability to detect/retrieve the payload on those modified watermarked NN.
  3. The computational cost of payload injection in the watermarked neural network, characterized by:
    1. The memory footprint.
    2. The time to execute the operation required by one epoch normalized according to the number of batches processed in one epoch.
    3. The time for the watermarked neural network to compute an inference.
    4. In case injection is done concurrently with network training, the number of epochs required to insert the watermark.
  4. The computational cost of payload detection or decoding from the watermarked neural network, characterized by:
    1. The memory footprint.
    2. The total duration.

Read about MPAI-NNW:

  1. 1 min 30 sec video (YouTube ) and video (non YouTube) illustrating MPAI-NNW V1.
  2. slides presented at the online meeting on 2022/07/12.
  3. video recording of the online presentation (Youtube, non-YouTube) made at that 12 July presentation
  4. Call for Technologies, the Use Cases and Functional Requirements, and the Framework Licence.

If you wish to participate in this work you have the following options

  1. Join MPAI
  2. Participate until the MPAI-NNW Functional Requirements are approved (after that only MPAI members can participate) by sending an email to the MPAI Secretariat.
  3. Keep an eye on this page.

Return to the MPAI-NNW page.