Cross-domain machine learning for imbalanced domains
Devices and techniques are generally described for cross-domain machine learning. A first machine learning model may be trained using first data of a first domain. Predictions may be generated by inputting a plurality of domain data from other domains apart from the first domain into the first machine learning model. For each of the predictions, a prediction error may be determined. A grouping of similar domains from among the other domains may be determined based on the prediction errors. A second machine learning model may be trained for the grouping of similar domains.
1. A computer-implemented method of cross-domain machine learning, the method comprising:
sending first data of a first domain to a first machine learning model trained using second data of a second domain;
determining a first error describing a difference between predicted values generated by the first machine learning model and actual values of the first data of the first domain;
sending the second data of the second domain to a second machine learning model trained using the first data of the first domain;
determining a second error describing a difference between predicted values generated by the second machine learning model and actual values of the second data of the second domain;
determining a similarity between the first domain and the second domain based on the first error and the second error;
training a third machine learning model using the first data of the first domain and the second data of the second domain;
receiving third data of the first domain; and
generating a prediction for the third data using the third machine learning model.
2. The computer-implemented method of claim 1 , further comprising:
generating, using the first error and the second error, a similarity matrix describing similarities among the first domain and the second domain.
3. The computer-implemented method of claim 1 , further comprising:
determining a plurality of groupings of domains; and
generating, for each of the plurality of groupings of domains, a respective fourth machine learning model.
4. The computer-implemented method of claim 1 , further comprising:
determining a first number of samples of the first data of the first domain;
determining a total number of samples across a plurality of domains including the first domain and the second domain; and
determining a sample weight using a ratio of the first number of samples to the total number of samples, wherein the training the third machine learning model comprises applying the sample weight to a loss function of the third machine learning model.
5. The computer-implemented method of claim 1 , further comprising:
determining a first optimized hyper-parameter for the first machine learning model; and
determining a second optimized hyper-parameter for the second machine learning model.
6. The computer-implemented method of claim 5 , further comprising:
training the first machine learning model using the first data of the first domain and the second data of the second domain; and
training the second machine learning model using the first data of the first domain and the second data of the second domain.
7. The computer-implemented method of claim 1 , further comprising:
receiving fourth data;
determining that the fourth data corresponds to the second domain;
sending the fourth data to the third machine learning model; and
generating, by inputting the fourth data into the third machine learning model, a prediction for the fourth data.
8. A system comprising:
at least one processor; and
non-transitory computer-readable memory storing instructions that, when executed by the at least one processor, are effective to:
send first data of a first domain to a first machine learning model trained using second data of a second domain;
determine a first error describing a difference between predicted values generated by the first machine learning model and actual values of the first data of the first domain;
send the second data of the second domain to a second machine learning model trained using the first data of the first domain;
determine a second error describing a difference between predicted values generated by the second machine learning model and actual values of the second data of the second domain;
determine a similarity between the first domain and the second domain based on the first error and the second error;
train a third machine learning model using the first data of the first domain and the second data of the second domain;
receive third data of the first domain; and
generate a prediction for the third data using the third machine learning model.
9. The system of claim 8 , the non-transitory computer-readable memory storing further instructions that, when executed by the at least one processor, are further effective to:
generate, using the first error and the second error, a similarity matrix describing similarities among the first domain and the second domain.
10. The system of claim 8 , the non-transitory computer-readable memory storing further instructions that, when executed by the at least one processor, are further effective to:
determine a plurality of groupings of domains; and
generate, for each of the plurality of groupings of domains, a respective fourth machine learning model.
11. The system of claim 8 , the non-transitory computer-readable memory storing further instructions that, when executed by the at least one processor, are further effective to:
determine a first number of samples of the first data of the first domain;
determine a total number of samples across a plurality of domains including the first domain and the second domain; and
determine a sample weight using a ratio of the first number of samples to the total number of samples, wherein the training the third machine learning model comprises applying the sample weight to a loss function of the third machine learning model.
12. The system of claim 8 , the non-transitory computer-readable memory storing further instructions that, when executed by the at least one processor, are further effective to:
determine a first optimized hyper-parameter for the first machine learning model; and
determine a second optimized hyper-parameter for the second machine learning model.
13. The system of claim 12 , the non-transitory computer-readable memory storing further instructions that, when executed by the at least one processor, are further effective to:
train the first machine learning model using the first data of the first domain and the second data of the second domain; and
train the second machine learning model using the first data of the first domain and the second data of the second domain.
14. The system of claim 8 , the non-transitory computer-readable memory storing further instructions that, when executed by the at least one processor, are further effective to:
receive fourth data;
determine that the fourth data corresponds to the second domain;
send the fourth data to the third machine learning model; and
generate, by inputting the fourth data into the third machine learning model, a prediction for the fourth data.
15. A computer-implemented method comprising:
sending first data of a first domain to a first machine learning model trained using second data of a second domain;
determining a first error describing a difference between predicted values generated by the first machine learning model and actual values of the first data of the first domain;
determining a similarity between the first domain and the second domain based on the first error and a second error associated with a second machine learning model trained using the first data of the first domain;
training a third machine learning model using the first data of the first domain and the second data of the second domain;
receiving third data of the first domain; and
generating a prediction for the third data using the third machine learning model.
16. The computer-implemented method of claim 15 , further comprising:
generating, using the first error and the second error, a similarity matrix describing similarities among the first domain and the second domain.
17. The computer-implemented method of claim 15 , further comprising:
determining a plurality of groupings of domains; and
generating, for each of the plurality of groupings of domains, a respective fourth machine learning model.
18. The computer-implemented method of claim 15 , further comprising:
determining a first number of samples of the first data of the first domain;
determining a total number of samples across a plurality of domains including the first domain and the second domain; and
determining a sample weight using a ratio of the first number of samples to the total number of samples, wherein the training the third machine learning model comprises applying the sample weight to a loss function of the third machine learning model.
19. The computer-implemented method of claim 15 , further comprising:
determining a first optimized hyper-parameter for the first machine learning model; and
determining a second optimized hyper-parameter for the second machine learning model.
20. The computer-implemented method of claim 19 , further comprising:
training the first machine learning model using the first data of the first domain and the second data of the second domain; and
training the second machine learning model using the first data of the first domain and the second data of the second domain.