Softmax

cosmos 4th November 2016 at 2:43pm

A Layers for deep learning, that consists on exponentiating all vector elements and normalize them to so sum to unity.

σ(y)eyieyi\mathbf{\sigma}(\mathbf{y}) \equiv \frac{e^\mathbf{y}}{\sum_i e^{y_i}}

where the mathematical operations are element-wise.