JPWO2021074770A5

JPWO2021074770A5 -

Info

Publication number: JPWO2021074770A5
Application number: JP2022521116A
Authority: JP
Publication date: 2022-12-22

Claims

A method for securing a trained machine learning model in a computing environment by one or more processors, the method comprising:
A method comprising providing one or more hardened machine learning models secured against adversarial attacks by adding adversarial protection to one or more trained machine learning models.

receiving the one or more trained machine learning models;
retraining the one or more trained machine learning models to include the adversarial protection based on one or more adversarial protection protocols;
2. The method of claim 1, further comprising:

3. The method of claim 1 or 2, further comprising adding one or more preprocessing layers to one or more trained machine learning models, wherein the degree of strength of adversarial protection is adjusted.

4. The method of any of claims 1-3, further comprising adding one or more neural network layers to one or more trained machine learning models.

further comprising applying one or more post-processing output layers to the one or more trained machine learning models, wherein the one or more trained machine learning models include one or more adversarial 5. A method according to any of claims 1 to 4, wherein retraining with said adversarial protection is based on a protection protocol.

automatically implementing one or more adversarial protection protocols used to provide said one or more enhanced machine learning models; or from a user, said one or more enhanced machine learning models. receiving one or more adversarial protection protocols used to provide learning models;
6. The method of any of claims 1-5, further comprising:

monitoring and tracking states of the one or more trained machine learning models while being retrained;
detecting training collapse for the one or more trained machine learning models during the retraining; or one or more roles for the one or more trained machine learning models during a retraining operation.・To enable the back strategy,
7. The method of any one of claims 1-6, further comprising:

A system for securing a trained machine learning model in a computing environment, comprising:
one or more computers having executable instructions that, when executed, cause the system to:
A system that provides one or more hardened machine learning models secured against adversarial attacks by adding adversarial protection to one or more trained machine learning models.

The executable instructions are:
receiving the one or more trained machine learning models;
retraining the one or more trained machine learning models to include the adversarial protection based on one or more adversarial protection protocols;
9. System according to claim 8.

10. The executable instructions of claim 8 or 9, wherein the executable instructions add one or more preprocessing layers to one or more trained machine learning models to adjust the degree of strength of adversarial protection. system.

11. The system of any of claims 8-10, wherein the executable instructions apply one or more neural network layers to one or more trained machine learning models.

The executable instructions apply one or more post-processing output layers to the one or more trained machine learning models, wherein the one or more trained machine learning models are adapted to one or more 12. The system according to any one of claims 8 to 11, retrained by said adversarial protection based on the adversarial protection protocol of .

The executable instructions are:
automatically implement one or more adversarial protection protocols used to provide said one or more enhanced machine learning models; or from a user, said one or more enhanced machine learning models. 13. The system of any of claims 8-12, receiving one or more adversarial protection protocols used to provide learning models.

The executable instructions are:
monitoring and tracking each state of the one or more trained machine learning models while being retrained;
detect training collapse for the one or more trained machine learning models during the retraining; or perform one or more rolls for the one or more trained machine learning models during a retraining operation. - A system according to any one of claims 8 to 13, which enables a back strategy.

A computer program product for causing a computer to perform the steps of the method according to any one of claims 1 to 7.

16. A computer readable medium storing the computer program according to claim 15.