Maxout Networks

Maxout is a kind of neural network activation function that I developed in January 2013. It works very well when trained with Geoff Hinton's dropout algorithm. I wrote an ICML paper about it with David Warde-Farley, Mehdi Mirza, Aaron Courville, and Yoshua Bengio. We used it to set the state-of-the-art on several benchmark tasks.

This is the project page for all things related to maxout. We might post future announcements related to it here.

Some resources related to maxout include: