Neural networks adaptive boosting using semi-supervised learning

1. A method for generating a trained neural network, comprising:

creating a neural network;

performing an initial training of the neural network using a set of labeled data;

performing a plurality of iterations in which a boosted neural network is applied to unlabeled data to determine whether any of the unlabeled data qualifies as additional labeled data;

retraining, in response to performing of an iteration in which any of the unlabeled data qualifies as additional labeled data, the boosted neural network using the additional labeled data; and

updating, in response to the performing of an iteration in which none of the unlabeled data qualifies as additional labeled data, the neural network to change a number of predictor nodes in the neural network, the updating further comprising removing at least one predictor node from a hidden layer of the neural network in response to a determination that the number of predictor nodes is greater than a predetermined number.

2. The method of claim 1 , wherein the neural network includes an input layer, an output layer, and a hidden layer.

3. The method of claim 2 , wherein the at least one predictor node includes a predictor node with a lowest filling rate.

4. The method of claim 2 , the updating further comprising: re-adding at least one node to the hidden layer of the neural network in response to a determination that the number of predictor nodes has reached a predetermined number.

5. The method of claim 4 , further comprising executing a final boosting on the neural network to finalize a structure of the neural network in response to the re-adding.

6. The method of claim 2 , wherein the initial training includes, for each labeled instance of the set of labeled data:

introducing the labeled instance at the input layer of the neural network;

evaluating the labeled instance by at least one node of the hidden layer;

outputting a solution at the output layer based on the evaluation;

comparing the solution with a labeled solution associated with the labeled instance; and

weighting the at least one node based on a result of the comparing to get the boosted neural network.

7. The method of claim 1 , further comprising:

determining, for each unlabeled instance of the unlabeled data, whether an outputted solution by the neural network is accurate;

labeling, in response to a determination that the solution is accurate, the unlabeled instance with the outputted solution to yield a labeled instance; and

adding the labeled instance to the set of labeled data.

8. A computer program product for generating a trained neural network, the computer program product comprising a computer readable storage media, and program instructions stored on the computer readable storage media, that cause at least one computer device to:

create a neural network;

perform an initial training of the neural network using a set of labeled data;

apply a boosted neural network to unlabeled data to determine whether any of the unlabeled data qualifies as additional labeled data;

retrain, in response to performing of an iteration in which any of the unlabeled data qualifies as additional labeled data, the boosted neural network using the additional labeled data; and

update, in response to the performing of an iteration in which none of the unlabeled data qualifies as additional labeled data, the neural network to change a number of predictor nodes in the neural network, the update further causing the at least one computer device to remove at least one predictor node from a hidden layer of the neural network in response to a determination that the number of predictor nodes is greater than a predetermined number.

9. The computer program product of claim 8 , wherein the neural network includes an input layer, an output layer, and a hidden layer.

10. The computer program product of claim 9 , wherein the at least one predictor node includes a predictor node with a lowest filling rate.

11. The computer program product of claim 9 , the instructions that cause the at least one computer device to update further causing the at least one computer device to re-add at least one node to the hidden layer of the neural network in response to a determination that the number of predictor nodes has reached a predetermined number.

12. The computer program product of claim 11 , the instructions further causing the at least one computer device to execute a final boosting on the neural network to finalize a structure of the neural network in response to the re-adding.

13. The computer program product of claim 9 , wherein the initial training includes, for each labeled instance of the set of labeled data:

introducing the labeled instance at the input layer of the neural network;

evaluating the labeled instance by at least one node of the hidden layer;

outputting a solution at the output layer based on the evaluation;

comparing the solution with a labeled solution associated with the labeled instance; and

weighting the at least one node based on a result of the comparing to get the boosted neural network.

14. The computer program product of claim 8 , the instructions further causing the at least one computer device to parse, prior to forming a machine language model, annotated documents to remove from a document unannotated portions of the document.

15. A system for generating a trained neural network, comprising:

a neural network having an input layer, an output layer, and a hidden layer;

a memory medium comprising instructions;

a bus coupled to the memory medium; and

a processor coupled to the bus that when executing the instructions causes the system to:

perform an initial training of the neural network using a set of labeled data;

apply a boosted neural network to unlabeled data to determine whether any of the unlabeled data qualifies as additional labeled data;

retrain, in response to performing of an iteration in which any of the unlabeled data qualifies as additional labeled data, the boosted neural network using the additional labeled data; and

update, in response to the performing of an iteration in which none of the unlabeled data qualifies as additional labeled data, the neural network to change a number of predictor nodes in the neural network, the update further causing the system to remove at least one predictor node from a hidden layer of the neural network in response to a determination that the number of predictor nodes is greater than a predetermined number.

16. The system of claim 15 , the instructions that cause the system to update further causing the system to re-add at least one node to the hidden layer of the neural network in response to a determination that the number of predictor nodes has reached a predetermined number, the at least one predictor node includes a predictor node with a lowest filling rate.

17. The system of claim 16 , the instructions further causing the at least one computer device to executing a final boosting on the neural network to finalize a structure of the neural network in response to the re-adding.

18. The system of claim 15 , wherein the initial training includes, for each labeled instance of the set of labeled data:

introducing the labeled instance at the input layer of the neural network;

evaluating the labeled instance by at least one node of the hidden layer;

outputting a solution at the output layer based on the evaluation;

comparing the solution with a labeled solution associated with the labeled instance; and

weighting the at least one node based on a result of the comparing to get the boosted neural network.

19. The system of claim 15 , the instructions further causing the system to train the artificial intelligence using a machine learning model.