Automatically labeling capability for training and validation data for machine learning

1. A method, comprising:

inputting candidate data into a first neural network to filter the candidate data by selecting a subset of the candidate data based on an output of the first neural network, wherein the first neural network is pretrained;

performing a confidence upgrade check on the subset of candidate data by:

performing a data consistency check by generating augmented data from each candidate data from among the subset of candidate data, wherein the generated augmented data are used as inputs into the first neural network to determine a first confidence condition for each of the subset of candidate data,

inputting the subset of candidate data into a second neural network that is trained using data from an environment to determine a second confidence condition, wherein the second neural network is a version of the first neural network overfitted to the environment, and

performing a clustering on the subset of candidate data, wherein results from the clustering on the subset of candidate data are used as inputs into a third machine learning approach to determine a third confidence condition; and

automatically labeling, as training data, the subset of candidate data from among the subset of candidate data in accordance with a confidence level label based on the first confidence condition, the second confidence condition, and the third confidence condition.

2. The method of claim 1 , wherein the candidate data corresponds to newly acquired data captured by sensors on a device, wherein the newly acquired data corresponds to visual or audio data specific to the environment.

3. The method of claim 1 , wherein the first neural network is based on a simplified version of a base neural network model.

4. The method of claim 1 , wherein performing the data consistency check further comprises:

assigning a group identification to the augmented data, wherein the group identification indicates a source for the generated augmented data;

comparing a candidate label with other candidate labels in an augmented data group with a same group identification; and

increasing a weight of the first confidence condition based on a confidence of the candidate label when the candidate label is consistent with the other candidate labels in the augmented data group with the same group identification.

5. The method of claim 1 , wherein the augmented data are created by at least one of: shifting, scaling, translating, or rotating an image from the subset of candidate data.

6. The method of claim 1 , wherein the clustering includes a known fixed number of clusters with known reference data points.

7. The method of claim 1 , wherein the clustering is based on comparing distance measurements for the subset of candidate data with known reference points and measuring similarities between the distance measurements in order to form clusters.

8. The method of claim 1 , wherein the first confidence condition is based on results from the data consistency check, the second confidence condition corresponds to a result from the output of the second neural network, and the third confidence condition corresponds to a result from the output of the third machine learning approach.

9. The method of claim 1 , wherein the automatically labeled subset of candidate data is differentiated into automatically-labeled training data and automatically-labeled validation data.

10. The method of claim 1 , wherein automatically labeling the subset of candidate data from among the subset of candidate data is performed while a device is charging at a dock station.

11. An edge device, the edge device comprising:

one or more processors;

a non-transitory memory; and

one or more programs stored in the non-transitory memory, which, when executed by the one or more processors, cause the edge device to be configured to perform:

inputting candidate data into a first neural network to filter the candidate data by selecting a subset of the candidate data based on an output of the first neural network, wherein the first neural network is pretrained;

performing a confidence upgrade check on the subset of candidate data by:

performing a data consistency check by generating augmented data from each candidate data from among the subset of candidate data, wherein the generated augmented data are used as inputs into the first neural network to determine a first confidence condition for each of the subset of candidate data,

inputting the subset of candidate data into a second neural network that is trained using data from an environment to determine a second confidence condition, wherein the second neural network is a version of the first neural network overfitted to the environment, and

performing a clustering on the subset of candidate data, wherein results from the clustering on the subset of candidate data are used as inputs into a third machine learning approach to determine a third confidence condition; and

automatically labeling, as training data, the subset of candidate data from among the subset of candidate data in accordance with a confidence level label based on the first confidence condition, the second confidence condition, and the third confidence condition.

12. The edge device of claim 11 , wherein the candidate data corresponds to newly acquired data captured by sensors on a device, wherein the newly acquired data corresponds to visual or audio data specific to the environment.

13. The edge device of claim 11 , wherein the first neural network is based on a simplified version of a base neural network model.

14. The edge device of claim 11 , wherein performing the data consistency check further comprises:

assigning a group identification to the augmented data, wherein the group identification indicates a source for the generated augmented data;

comparing a candidate label with other candidate labels in an augmented data group with a same group identification; and

increasing a weight of the first confidence condition based on a confidence of the candidate label when the candidate label is consistent with the other candidate labels in the augmented data group with the same group identification.

15. The edge device of claim 11 , wherein the augmented data are created by at least one of: shifting, scaling, translating, or rotating an image from the subset of candidate data.

16. The edge device of claim 11 , wherein the clustering includes a known fixed number of clusters with known reference points.

17. The edge device of claim 11 , wherein the clustering is based on comparing distance measurements for the subset of candidate data with known reference points and measuring similarities between the distance measurements in order to form clusters.

18. The edge device of claim 11 , wherein the first confidence condition is based on results from an inference result consistency check, the second confidence condition corresponds to a weighted sum of an output from the second neural network, and the third confidence condition corresponds to a result from the output of the third machine learning approach.

19. The edge device of claim 11 , wherein the automatically labeled subset of candidate data is differentiated into automatically-labeled training data and automatically-labeled validation data.

20. The edge device of claim 11 , wherein automatically labeling the subset of candidate data from among the subset of candidate data is performed while the edge device is charging at a dock station.

21. A machine-readable non-transitory medium having stored thereon machine-executable instructions for labeling training data for machine learning, the instructions comprising:

inputting candidate data into a first neural network to filter the candidate data by selecting a subset of the candidate data based on an output of the first neural network, wherein the first neural network is pretrained;

performing a confidence upgrade check on the subset of candidate data by:

performing a data consistency check by generating augmented data from each candidate data from among the subset of candidate data, wherein the generated augmented data are used as inputs into the first neural network to determine a first confidence condition for each of the subset of candidate data,

inputting the subset of candidate data into a second neural network that is trained using data from an environment to determine a second confidence condition, wherein the second neural network is a version of the first neural network overfitted to the environment, and

performing a clustering on the subset of candidate data, wherein results from the clustering on the subset of candidate data are used as inputs into a third machine learning approach to determine a third confidence condition; and

automatically labeling, as training data, the subset of candidate data from among the subset of candidate data in accordance with a confidence level label based on the first confidence condition, the second confidence condition, and the third confidence condition.