Activation Functions
Non-linear functions in neural networks
Allow strong signals through, block weak signals
Types:
- sigmoid - binary classification
- relu - image classification
- softmax - multi-class output
- tanh - NLP tasks
Provides non-linearity for artificial-neural-network
→ Enables learning complex patterns → Types: ReLU, sigmoid, tanh, softmax → Without it, network collapses to linear regression
References
- Activation Functions in Neural Networks
- Activation Functions Guide - DeepAI
- Understanding Activation Functions - Towards Data Science #ml-notes
- Comparative Study of Activation Functions
#ml-notes