Towards A Differential Privacy and Utility Preserving Machine Learning Classifier

What is it about?

Because of too much noise generated during the privacy process, privatized data sets tend to have lower usability (utility). One way to deal with this problem is to use machine learning as a gauge for better usability of the privatized data. In the paper we apply data privacy on a data set using differential privacy, a state-of-the-art data privacy technique. We then apply machine learning on the privatized data set to gauge the usability of that data.

Featured Image

Why is it important?

Finding a balance between privacy and usability has been found to be intractable. There is always a tension between publishing a fully privatized data set but such data will have low utility. Therefore, a balance is always sought but with trade-offs. Meaning that you might achieve an acceptable amount of utility but at the cost of lower privacy, or versa-visa.

Perspectives