Identification of copied ML model

Patent metadata

Field	Value
Publication number	US-12112246-B2
Application number	US-201817281602-A
Country	US
Kind code	B2
Filing date	Oct 1, 2018
Priority date	Oct 1, 2018
Publication date	Oct 8, 2024
Grant date	Oct 8, 2024

Abstract

There is provided mechanisms for a manufacturer of an ML model to embed at least one marker in an electronic file. A method comprises obtaining the electronic file. The electronic file represents content that causes the ML model to determine an output for the electronic file according to a first processing strategy. The method comprises embedding, in the electronic file, the at least one marker that, only when detected by the ML model, causes the output of the electronic file to be determined according to a second processing strategy. The second processing strategy is unrelated to the first processing strategy and deterministically defined by the at least one marker.

First claim

The invention claimed is:

1. A method for a manufacturer of a machine learning (ML) model to embed at least one marker in an electronic file, the method comprising:
providing an initial electronic file as input to an initial ML model for processing, wherein content of the initial electronic file is defined by the at least one marker, and wherein an output of the initial electronic file is determined by the initial ML model according to an initial processing strategy, the output being exclusive only for electronic files comprising the at least one marker;
feeding the output of the initial electronic file as input to a reverse classifier that generates a second electronic file as its output;
obtaining the second electronic file, wherein the second electronic file represents content that causes a second ML model to determine an output for the second electronic file according to a first processing strategy; and
embedding, in the second electronic file, the at least one marker that, only when detected by the second ML model, causes the output of the second electronic file to be determined according to a second processing strategy, the second processing strategy being unrelated to the first processing strategy and deterministically defined by the at least one marker.

2. The method of claim 1 :
wherein to determine the output for the second electronic file according to the first processing strategy comprises classifying the second electronic file into a first classification category; and
wherein to determine the output for the second electronic file according to the second processing strategy comprises classifying the second electronic file into a second classification category.

3. The method of claim 2 , wherein the second classification category is an identifier of an original manufacturer of the ML model.

4. The method of claim 3 , wherein the identifier is a telephone number, name of the manufacturer, or address of the manufacturer.

5. The method of claim 1 :
wherein to determine the output for the second electronic file according to the first processing strategy comprises making a prediction on data of the second electronic file according to a first predictor; and
wherein to determine the output for the second electronic file according to the second processing strategy comprises making the prediction on the data of the second electronic file according to a second predictor.

6. The method of claim 1 , wherein the second ML model has been trained to, based on the content, determine the output for the second electronic file according to the first processing strategy.

7. The method of claim 1 , wherein the second ML model has been trained to:
detect the at least one marker; and
determine, upon detection of the at least one marker, the output for the second electronic file according to the second processing strategy.

8. The method of claim 1 , wherein the second processing strategy, except for the at least one marker, is unrelated to the content represented by the second electronic file.

9. The method of claim 1 , wherein the at least one marker is a steganographic marker.

10. The method of claim 1 , wherein the second electronic file is processed according to the second processing strategy only when the at least one marker has a predetermined characteristic in the second electronic file and/or a predetermined relation to content of the second electronic file.

11. The method of claim 10 :
wherein each of the at least one marker is selected from a set of markers of different types;
wherein the following represent values that are given as input to a triggering function:
which type of marker each of the at least one marker is; and
a location, size, rotation, and/or transformation of each marker in the second electronic file; and/or
a predetermined relation of each marker to content of the second electronic file;

wherein the second ML model computes a value of the triggering function during processing of the second electronic file; and
wherein the output for the second electronic file is determined according to the second processing strategy only when the triggering function is computed to a predetermined value.

12. The method of claim 1 , wherein, when there are at least two markers, the second electronic file is processed according to the second processing strategy only when the at least two markers have a predetermined relation in the second electronic file.

13. The method of claim 1 , wherein the at least one marker is represented by how the content is structured in the second electronic file.

14. The method of claim 1 , wherein at least one further marker is embedded in the second electronic file before the second electronic file is provided as input to the second ML model for classification.

15. The method of claim 1 , wherein the content represents an image, audio, video, a document, traffic data, and/or weather data.

16. A method for identifying whether a machine learning (ML) model belongs to a manufacturer of the ML model or not, the method comprising:
providing an initial electronic file as input to an initial ML model for processing, wherein content of the initial electronic file is defined by at least one marker, and wherein an output of the initial electronic file is determined by the initial ML model according to an initial processing strategy, the output being exclusive only for electronic files comprising the at least one marker;
feeding the output of the initial electronic file as input to a reverse classifier that generates a second electronic file as its output;
providing the second electronic file as input to a second ML model for processing, wherein the second electronic file represents content that causes the second ML model to determine an output for the second electronic file according to a first processing strategy, and wherein the second electronic file comprises the at least one marker embedded in the electronic file by the manufacturer and that, only when detected by the second ML model, causes the output of the second electronic file to be determined according to a second processing strategy, the second processing strategy being unrelated to the first processing strategy and deterministically defined by the at least one marker; and
identifying the second ML model as belonging to the manufacturer only when the output of the second electronic file is by the second ML model determined according to the second processing strategy.

17. An electronic device for a manufacturer of a machine learning (ML) model to embed at least one marker in an electronic file, the electronic device comprising:
processing circuitry;
memory containing instructions executable by the processing circuitry whereby the electronic device is operative to:
provide an initial electronic file as input to an initial ML model for processing, wherein content of the initial electronic file is defined by the at least one marker, and wherein an output of the initial electronic file is determined by the initial ML model according to an initial processing strategy, the output being exclusive only for electronic files comprising the at least one marker;
feed the output of the initial electronic file as input to a reverse classifier that generates a second electronic file as its output;
obtain the second electronic file, wherein the second electronic file represents content that causes a second ML model to determine an output for the second electronic file according to a first processing strategy; and
embed, in the second electronic file, the at least one marker that, only whe

Assignees

• Ericsson Telefon Ab L M

Inventors

• Slavov Kristian
• Salmela Patrik
• Ramos Edgar

Classifications

• G06F21/16

  Program or content traceability, e.g. by watermarking · CPC title

• G06N20/00 Primary

  Machine learning · CPC title