Self-supervised self supervision by combining probabilistic logic with deep learning
The present disclosure relates to devices and methods for determining new virtual evidence to use with a deep probabilistic logic module. The devices and methods may receive output from a deep probabilistic logic module in response to running an initial set of virtual evidence through the deep probabilistic logic module. The devices and methods may use the output to automatically propose at least one factor as new virtual evidence for use with the deep probabilistic logic module. The devices and methods may add the new virtual evidence to the deep probabilistic logic module.
1. A method, comprising:
creating, using an initial set of virtual evidence, a factor graph to provide to a deep probabilistic logic module;
adjusting values of the virtual evidence in the factor graph in response to recommendations from the deep probabilistic logic module; and
using the factor graph to train a deep neural network to predict labels for data.
2. The method of claim 1 , further comprising:
generating, using the factor graph and the labels for the data, new virtual evidence to add to the factor graph.
3. The method of claim 2 , further comprising:
retraining, using the factor graph with the initial set of virtual evidence and the new virtual evidence, the deep neural network to predict the labels for the data.
4. The method of claim 3 , further comprising:
refining virtual evidence parameters in response to retraining of the deep neural network.
5. The method of claim 4 , wherein refining the virtual evidence parameters includes changing a label of existing virtual evidence or modifying weights of existing virtual evidence.
6. The method of claim 1 , further comprising:
resolving a difference in a prediction for variables in the virtual evidence and a latent variable label for the variables in the virtual evidence; and
providing the recommendations to update the values of the virtual evidence.
7. The method of claim 1 , wherein adjusting the values of the virtual evidence further includes continuing to adjust the values of the virtual evidence until the factor graph of the deep probabilistic logic module and the deep neural network agree on a prediction for the initial set of the virtual evidence.
8. The method of claim 1 , wherein the initial set of virtual evidence includes words and associated sentiment labels.
9. The method of claim 1 , wherein the initial set of virtual evidence includes noisy virtual evidence that defines a probabilistic distribution of variables over unknown latent variable labels.
10. The method of claim 1 , wherein the factor graph includes unknown latent variable labels for the data.
11. The method of claim 10 , wherein the virtual evidence includes a confidence in an estimated distribution of the data over the unknown latent variable labels.
12. A device, comprising:
a memory to store data and instructions; and
a processor operable to communicate with the memory, wherein the processor is operable to:
create, using an initial set of virtual evidence, a factor graph to provide to a deep probabilistic logic module;
adjust values of the virtual evidence in the factor graph in response to recommendations from the deep probabilistic logic module; and
use the factor graph to train a deep neural network to predict labels for data.
13. The device of claim 12 , wherein the processor is further operable to:
generate, using the factor graph and the labels for the data, new virtual evidence to add to the factor graph.
14. The device of claim 13 , wherein the processor is further operable to:
retrain, using the factor graph with the initial set of virtual evidence and the new virtual evidence, the deep neural network to predict the labels for the data.
15. The device of claim 14 , wherein the processor is further operable to:
refine virtual evidence parameters in response to the retraining of the deep neural network.
16. The device of claim 15 , wherein the processor is further operable to refine the virtual evidence parameters by changing a label of existing virtual evidence or modifying weights of existing virtual evidence.
17. The device of claim 12 , wherein the processor is further operable to:
resolve a difference in a prediction for variables in the virtual evidence and a latent variable label for the variables in the virtual evidence; and
provide the recommendations to update the values of the virtual evidence.
18. The device of claim 12 , wherein the processor is further operable to adjust the values of the virtual evidence by continuing to adjust the values of the virtual evidence until the factor graph of the deep probabilistic logic module and the deep neural network agree on a prediction for the initial set of the virtual evidence.
19. The device of claim 12 , wherein the factor graph includes unknown latent variable labels for the data.
20. The device of claim 19 , wherein the virtual evidence includes a confidence in an estimated distribution of the data over the unknown latent variable labels.