Distributed labeling for supervised learning
Embodiments described herein provide a technique to crowdsource labeling of training data for a machine learning model while maintaining the privacy of the data provided by crowdsourcing participants. Client devices can be used to generate proposed labels for a unit of data to be used in a training dataset. One or more privacy mechanisms are used to protect user data when transmitting the data to a server. The server can aggregate the proposed labels and use the most frequently proposed labels for an element as the label for the element when generating training data for the machine learning model. The machine learning model is then trained using the crowdsourced labels to improve the accuracy of the model.
1. A method comprising:
receiving, by an electronic device, an unlabeled set of data from a server;
generating, by the electronic device and using a trained machine learning model, proposed labels for elements of the unlabeled set of data; and
transmitting, by the electronic device, a privatized version of one of the proposed labels to the server.
2. The method of claim 1 , further comprising:
generating, by the electronic device, the privatized version of the one of the proposed labels via a privacy-preserving encoding.
3. The method of claim 1 , wherein the unlabeled set of data includes one or more of text data, image data, application activity data, and device activity data.
4. The method of claim 1 , wherein the trained machine learning model is configured as a convolutional neural network.
5. The method of claim 1 , wherein the trained machine learning model is configured as a recurrent neural network.
6. The method of claim 1 , wherein the proposed labels are encoded to mask individual contributors via a privacy-preserving encoding algorithm.
7. The method of claim 6 , wherein the privacy-preserving encoding algorithm is a differential privacy algorithm.
8. The method of claim 1 , further comprising:
processing the proposed labels to determine an estimate of a most frequent proposed label for the unlabeled set of data by generating a sketch matrix of received proposed labels to aggregate proposed label data.
9. The method of claim 1 , further comprising:
processing the proposed labels to determine an estimate of a most frequent proposed label for the unlabeled set of data by generating a histogram of received proposed labels to aggregate proposed label data.
10. A device comprising:
a memory; and
at least one processor configured to:
receive an unlabeled set of data from a server;
generate, using a trained machine learning model, proposed labels for elements of the unlabeled set of data; and
transmit a privatized version of one of the proposed labels to the server.
11. The device of claim 10 , wherein the at least one processor is further configured to:
generate the privatized version of the one of the proposed labels via a privacy-preserving encoding.
12. The device of claim 10 , wherein the unlabeled set of data includes one or more of text data, image data, application activity data, and device activity data.
13. The device of claim 10 , wherein the trained machine learning model is configured as at least one of a convolutional neural network or a recurrent neural network.
14. The device of claim 10 , wherein the proposed labels are encoded to mask individual contributors via a privacy-preserving encoding algorithm.
15. The device of claim 14 , wherein the privacy-preserving encoding algorithm is a differential privacy algorithm.
16. A non-transitory machine-readable medium comprising instructions that, when executed by one or more processors, cause the one or more processors to perform operations comprising:
receiving, by an electronic device, an unlabeled set of data from a server;
generating, by the electronic device and using a trained machine learning model, proposed labels for elements of the unlabeled set of data; and
transmitting, by the electronic device, a privatized version of one of the proposed labels to the server.
17. The non-transitory machine-readable medium of claim 16 , wherein the operations further comprise:
generating, by the electronic device, the privatized version of the one of the proposed labels via a privacy-preserving encoding.
18. The non-transitory machine-readable medium of claim 16 , wherein the unlabeled set of data includes one or more of text data, image data, application activity data, and device activity data.
19. The non-transitory machine-readable medium of claim 16 , wherein the trained machine learning model is configured as a convolutional neural network.
20. The non-transitory machine-readable medium of claim 16 , wherein the proposed labels are encoded to mask individual contributors via a privacy-preserving encoding algorithm.