r/MachineLearning - [R] New activation function - Piecewise Linear Unit - better than ReLU

Here's a little write-up on it, you can skim through the pretty pictures: Give it a try on some models and let me know how it works for you.