#sigmoid-functions

Sigmoid functions saturate and kill gradients — use ReLU instead

Sigmoid activation functions hinder neural network training by saturating, causing vanishing gradients; modern architectures favor ReLU and its variants for better performance [Astral Codex Ten].