dawg / models

Machine learning models for use in Vusic
8 stars 0 forks source link

Regularization #31

Closed aodonnell closed 5 years ago

aodonnell commented 5 years ago

Current Behaviour

Output of our masker is very prone to overfitting since some chunks produce a loss of zero

Suggested Fix

Add regularization to the output of the masker to supplement lost information and improve the duration of temporal context built up in the system.