# Softmax The softmax function takes a vector $x$ of $k$ real numbers, and normalizes it into a *probability distribution*: The standard softmax is defined as: $\sigma(\mathbf{x})_i = \frac{e^{\beta x_i}}{\sum_{j=1}^k e^{\beta_j x_j}} $ A nice intuition is that the softmax can be thought of as a **maximum** function, defined in a **differentiable** way. --- Date: 20221010 Links to: [Neural Networks MOC](Neural%20Networks%20MOC.md) Tags: #review References: * []()