Systems and methods for learning for domain adaptation

1. A method for training parameters of a first domain adaptation model, comprising:

evaluating a cycle consistency objective using a first task specific model associated with a first domain and a second task specific model associated with a second domain based on:

one or more first training representations adapted from the first domain to the second domain by a first domain adaptation model and from the second domain to the first domain by a second domain adaptation model; and

one or more second training representations adapted from the second domain to the first domain by the second domain adaptation model and from the first domain to the second domain by the first domain adaptation model;

evaluating one or more first discriminator models to generate a first discriminator objective using the second task specific model,

wherein the one or more first discriminator models include a plurality of discriminators corresponding to a plurality of bands that corresponds domain variable ranges of the first and second domains respectively;

evaluating a learning objective based on the cycle consistency objective and the first discriminator objective; and

updating one or more parameters of the first domain adaptation model based on the learning objective.

2. The method of claim 1 , wherein the cycle consistency objective includes:

a first task specific loss function associated with the first task specific model; and

a second task specific loss function associated with the second task specific model.

3. The method of claim 1 , further comprising:

evaluating the one or more first discriminator models to generate the first discriminator objective using the second task specific model based on:

one or more third training representations adapted from the first domain to the second domain by the first domain adaptation model, and

one or more fourth training representations in the second domain.

4. The method of claim 3 , further comprising:

evaluating one or more second discriminator models to generate a second discriminator objective using the first task specific model based on:

one or more fifth training representations adapted from the second domain to the first domain by the second domain adaptation model, and

one or more sixth training representations in the first domain;

wherein the evaluating the learning objective includes:

evaluating the learning objective based on the cycle consistency objective and first and second discriminator objectives.

5. The method of claim 3 , wherein each of the plurality of discriminators is configured to discriminate between the adapted third training representations and representations in the second domain.

6. The method of claim 5 , wherein the one or more first discriminator models include:

a first-band discriminator corresponding to a first band of the plurality of bands having a first width of the domain variable, and

a second-band discriminator corresponding to a second band of the plurality of bands having a second width of the domain variable different from the first width.

7. The method of claim 1 , wherein the first domain includes both labeled samples and unlabeled samples.

8. The method of claim 1 , wherein a supervised task model of the first task specific model includes an image recognition task model, an image segmentation task model, a semantic segmentation task model, a speech recognition task model, or a machine translation task model.

9. The method of claim 1 , wherein an unsupervised task of the first task specific model includes a video prediction task model, an object tracking task model, a language modeling task model, or a speech modeling task model.

10. The method of claim 1 , wherein the second domain includes at least one labeled sample and at least one unlabeled sample.

11. A non-transitory machine-readable medium comprising a plurality of machine-readable instructions which, when executed by one or more processors, are adapted to cause the one or more processors to perform a method comprising:

evaluating a cycle consistency objective using a first task specific model associated with a first domain and a second task specific model associated with a second domain based on:

one or more first training representations adapted from the first domain to the second domain by a first domain adaptation model and from the second domain to the first domain by a second domain adaptation model; and

one or more second training representations adapted from the second domain to the first domain by the second domain adaptation model and from the first domain to the second domain by the first domain adaptation model;

evaluating one or more first discriminator models to generate a first discriminator objective using the second task specific model,

wherein the one or more first discriminator models include a plurality of discriminators corresponding to a plurality of bands;

evaluating a learning objective based on the cycle consistency objective and the first discriminator objective; and

updating one or more parameters of the first domain adaptation model based on the learning objective.

12. The non-transitory machine-readable medium of claim 11 , wherein the cycle consistency objective includes:

a first task specific loss function associated with the first task specific model; and

a second task specific loss function associated with the second task specific model.

13. The non-transitory machine-readable medium of claim 11 , wherein the method further comprises:

evaluating the one or more first discriminator models to generate the first discriminator objective using the second task specific model based on:

one or more third training representations adapted from the first domain to second domain by the first domain adaptation model, and

one or more fourth training representations in the second domain.

14. The non-transitory machine-readable medium of claim 13 , wherein each of the plurality of bands corresponds to a domain variable range of a domain variable of the first and second domains, and

wherein each of the plurality of discriminators is configured to discriminate between the adapted third training representations and representations in the second domain.

15. The non-transitory machine-readable medium of claim 11 , wherein the first domain includes both labeled samples and unlabeled samples.

16. A system, comprising:

a non-transitory memory; and

one or more hardware processors coupled to the non-transitory memory and configured to read instructions from the non-transitory memory to cause the system to perform a method comprising:

evaluating a cycle consistency objective using a first task specific model associated with a first domain and a second task specific model associated with a second domain based on:

one or more first training representations adapted from the first domain to the second domain by a first domain adaptation model and from the second domain to the first domain by a second domain adaptation model; and

one or more second training representations adapted from the second domain to the first domain by the second domain adaptation model and from the first domain to the second domain by the first domain adaptation model;

evaluating one or more first discriminator models to generate a first discriminator objective using the second task specific model,

wherein the one or more first discriminator models include a plurality of discriminators corresponding to a plurality of bands;

evaluating a learning objective based on the cycle consistency objective and the first discriminator objective; and

updating one or more parameters of the first domain adaptation model based on the learning objective.

17. The system of claim 16 , wherein the cycle consistency objective includes:

a first task specific loss function associated with the first task specific model; and

a second task specific loss function associated with the second task specific model.

18. The system of claim 16 , wherein the method further comprises:

evaluating the one or more first discriminator models to generate the first discriminator objective using the second task specific model based on:

one or more third training representations adapted from the first domain to the second domain by the first domain adaptation model, and

one or more fourth training representations in the second domain.

19. The system of claim 18 , wherein each of the plurality of bands corresponds to a domain variable range of a domain variable of the first and second domains, and

wherein each of the plurality of discriminators is configured to discriminate between the adapted third training representations and representations in the second domain.

20. The system of claim 16 , wherein the second domain includes at least one labeled sample and at least one unlabeled sample.