Refactor how the dataset is collected for the Gaussian mapping
Closed this issue · 0 comments
AlstonLin commented
Right now, the variables (sigma and mu) are updated only when the user enters the pattern wrong.
This is problematic because it eventually makes the variance (sigma) too large and make it easy for imposters to mimic the owner.
Instead, the training set should update every single the the user successfully unlocks the phone without a prompt if they got in without the PIN, or with it if they needed to use the PIN.