What is it?
Used for Deep Learning mainly as an activation function of an Artificial Neural Network, the Heaviside step function, also called as unit step function, denoted by or , normally outputs 0 for negative arguments, and outputs 1 for positive arguments. Different conventions may propose that yield different values, but this is normally treated as the Sign Function.
The Heaviside function can be mathematically written as:
1= x \geq 0 \\ 0=x < 0 \end{cases}$$ Which would yield the following function: ![[step function.png|600]] ___ # Limitations Because of its binary nature, the *Heaviside function* may **not be the best option for problems where flexibility is needed**. For [[Deep Learning]] purposes, most cases needs models to have greater flexibility to be able to learn from raw data and prevent [[Underfitting]]. It also does not introduce *non-linearity*, which for complex problems with *non-linear* decision boundaries or complicated data, **will not be enough**. For these cases, other *activation functions* are recommended, like the [[Sigmoid Function]], [[Tanh Function]], and [[ReLU Function]]