Introspective extraction and complement control
A method and system of training a natural language processing network are provided. A corpus of data is received and one or more input features selected therefrom by a generator network. The one or more selected input features from the generator network are received by a first predictor network and used to predict a first output label. A complement of the selected input features from the generator network are received by a second predictor network and used to predict a second output label.
1. A computer implemented system for training a natural language processing network, comprising:
a generator network operative to receive a corpus of data and select one or more input features from the corpus of data;
a first predictor network operative to receive the one or more selected input features from the generator network and predict a first output label based on the received one or more selected input features; and
a second predictor network operative to receive a complement of the selected input features from the generator network and predict a second output label based on the received complement of the selected input features.
2. The system of claim 1 , wherein the generator network is configured to play an adversarial game with the second predictor network to make the second predictor network as ineffective to predict an output similar to that of the first predictor network, as possible.
3. The system of claim 1 , wherein the generator network is configured to play a minimax game with the second predictor network to make the second predictor network as ineffective to predict an output similar to that of the first predictor network, as possible.
4. The system of claim 1 , wherein the complement of the selected input features is based on one or more input features not selected by the generator network for the first predictor network.
5. The system of claim 1 , wherein the computer is configured to:
compare the first output label to the second output label; and
upon determining that the first output label is within a predetermined threshold from the second output label, adjusting the generator network to change a selection of the one or more input features from the corpus of data.
6. The system of claim 5 , wherein adjusting the generator network to change a selection of the one or more input features from the corpus of data comprises: including input features from the complement of the selected input features from the generator network.
7. The system of claim 5 , wherein the generator network is iteratively adjusted until the first output label is outside a predetermined second threshold from the second output label.
8. The system of claim 1 , wherein the first and second output labels are binary.
9. The system of claim 1 , wherein the training is unsupervised.
10. The system of claim 1 , wherein the number of input features identified by the generator network is limited based on a computational capability of the computer.
11. The system of claim 1 , wherein the generator network is an introspective generator that predicts a label before selecting the one or more input features from the corpus of data.
12. The system of claim 11 , wherein the generator network has a classifier having an architecture that is similar to that of the first predictor network.
13. A computing device comprising:
a processor;
a storage device coupled to the processor;
a program stored in the storage device, wherein an execution of the program by the processor configures the computing device to perform acts comprising:
receiving a corpus of data by a generator network;
selecting one or more input features from the corpus of data by the generator network;
receiving the one or more selected input features from the generator network by a first predictor network;
predicting a first output label by the first predictor network based on the received one or more selected input features;
receiving a complement of the selected input features from the generator network by a second predictor network; and
predicting a second output label by the second predictor network, based on the received complement of the selected input features.
14. The computing device of claim 13 , wherein the generator network is configured to play an adversarial game with the second predictor network to make the second predictor network as ineffective to predict an output similar to that of the first predictor network, as possible.
15. The computing device of claim 13 , wherein the complement of the selected input features is based on one or more input features not selected by the generator network for the first predictor network.
16. The computing device of claim 13 , wherein execution of the program by the processor further configures the computing device to perform acts comprising:
comparing the first output label to the second output label; and
upon determining that the first output label is within a predetermined threshold from the second output label, adjusting the generator network to change a selection of the one or more input features from the corpus of data by including input features from the complement of the selected input features from the generator network.
17. The computing device of claim 16 , wherein the generator network is iteratively adjusted until the first output label is outside a predetermined second threshold from the second output label.
18. The computing device of claim 13 , wherein the generator network is an introspective generator that predicts a label before selecting the one or more input features from the corpus of data.
19. A non-transitory computer readable storage medium tangibly embodying a computer readable program code having computer readable instructions that, when executed, causes a computer device to carry out a method of training a natural language processing network, comprising:
receiving a corpus of data by a generator network;
selecting one or more input features from the corpus of data by the generator network;
receiving the one or more selected input features from the generator network by a first predictor network;
predicting a first output label by the first predictor network based on the received one or more selected input features;
receiving a complement of the selected input features from the generator network by a second predictor network; and
predicting a second output label by the second predictor network, based on the received complement of the selected input features.
20. The non-transitory computer readable storage medium of claim 19 , wherein the generator network is configured to play an adversarial game with the second predictor network to make the second predictor network as ineffective to predict an output similar to that of the first predictor network, as possible.
21. The non-transitory computer readable storage medium of claim 19 , wherein the complement of the selected input features is based on one or more input features not selected by the generator network for the first predictor network.
22. The non-transitory computer readable storage medium of claim 19 , further comprising:
comparing the first output label to the second output label; and
upon determining that the first output label is within a predetermined threshold from the second output label, adjusting the generator network to change a selection of the one or more input features from the corpus of data by including input features from the complement of the selected input features from the generator network.
23. The non-transitory computer readable storage medium of claim 22 , wherein the generator network is iteratively adjusted until the first output label is outside a predetermined second threshold from the second output label.
24. The non-transitory computer readable storage medium of claim 19 , wherein the generator network is an introspective generator that predicts a label before selecting the one or more input features from the corpus of data.