Noise injection is a technique used in machine learning and deep learning models that works to protect online privacy by including random noise to the input data during training. Researchers at University of Ottawa studied tracking and profiling techniques used by companies, and in response developed a privacy app that injected noise based on genuine user activity to delude tracking tools.
The app was tested against various websites that track their users and the impact was measured through the analysis of the advertisements that appear on those websites. Researchers found that the correlation between the advertisements and previous user activity significantly decreased after the use of the privacy app.
- Check out the project presentation: Contextual Noise Injection for Online Privacy