activation function

This term is a technical specification within the field of machine learning and computational neuroscience. It describes the mechanism that allows a neural network to learn complex, non-linear relationships in data, preventing the entire network from behaving like a single linear regression model.

In a practical implementation context, the choice of function—such as ReLU, Sigmoid, or Tanh—significantly impacts the model's convergence speed and its susceptibility to issues like the vanishing gradient problem. It is almost exclusively used in technical documentation, academic papers, and software engineering discussions regarding AI.