View source for 17915336. NEURAL NETWORK WATERMARKING simplified abstract (Telefonaktiebolaget LM Ericsson (publ))

=NEURAL NETWORK WATERMARKING=


==Organization Name==

[[:Category:Telefonaktiebolaget LM Ericsson (publ)|Telefonaktiebolaget LM Ericsson (publ)]]


[[Category:Telefonaktiebolaget LM Ericsson (publ)]]

==Inventor(s)==

[[:Category:Jakob Sternby of Lund (SE)|Jakob Sternby of Lund (SE)]][[Category:Jakob Sternby of Lund (SE)]]

[[:Category:Björn Johansson of Bjärred (SE)|Björn Johansson of Bjärred (SE)]][[Category:Björn Johansson of Bjärred (SE)]]

==NEURAL NETWORK WATERMARKING - A simplified explanation of the abstract==


This abstract first appeared for US patent application 17915336 titled 'NEURAL NETWORK WATERMARKING

==Simplified Explanation==

The abstract describes a method for training a neural network and embedding a watermark in the network to prove ownership. The network has two sets of trainable parameters, and the method involves training the network with a first set of samples while hindering the update of the second set of parameters. Then, a second set of samples is used to embed the watermark by updating the second set of parameters while hindering the update of the first set of parameters.

* The method involves training a neural network and embedding a watermark in it.
* The network has two sets of trainable parameters.
* The first set of parameters is updated during the training process with a first set of samples.
* The second set of parameters is hindered from being updated during the training process with the first set of samples.
* The watermark is embedded by updating the second set of parameters with a second set of samples.
* The first set of parameters is hindered from being updated during the embedding process with the second set of samples.

== Potential Applications ==
* Intellectual property protection: The embedded watermark can be used to prove ownership of the neural network, helping to protect against unauthorized use or copying.
* Copyright protection: The watermark can be used to identify and track copyrighted neural networks, ensuring proper attribution and preventing infringement.
* Authentication: The embedded watermark can serve as a form of authentication, allowing the verification of the network's origin and integrity.

== Problems Solved ==
* Ownership proof: The embedded watermark provides a means to prove ownership of a neural network, addressing the challenge of intellectual property protection.
* Unauthorized use prevention: By embedding a watermark, the method helps prevent unauthorized use or copying of the neural network, addressing the challenge of copyright protection.
* Authentication assurance: The watermark allows for the authentication of the network's origin and integrity, addressing the challenge of ensuring trustworthiness in neural networks.

== Benefits ==
* Enhanced intellectual property protection: The embedded watermark provides a tangible proof of ownership, making it easier to enforce intellectual property rights.
* Deterrence against unauthorized use: The presence of a watermark can act as a deterrent, discouraging unauthorized use or copying of the neural network.
* Trust and integrity assurance: The embedded watermark allows for the verification of the network's origin and integrity, ensuring trustworthiness in its use.


==Original Abstract Submitted==

Training a neural network and embedding a watermark in the network to prove ownership. The network includes a plurality of trainable parameters associated with network nodes in which the plurality of trainable parameters is split into a first set of trainable parameters and a second set of trainable parameters. A first set of training samples is input to the network and the network is trained by iterating the first set of samples through the network to update the first set of parameters and hindering the second set of parameters to be updated during iteration of the first set of samples. A second set of samples is input and the watermark is embedded by iterating the second set of samples through the network to update the second set of parameters and hindering the first set of parameters to be updated during iteration of the second set of samples.



[[Category:G06F21/16]]
[[Category:G06N3/08]]