What are some common types of activation functions in neural networks, and what are their characteristics?

Common activation functions include: ReLU (Rectified Linear Unit), which introduces non-linearity and is computationally efficient; Sigmoid, which outputs values between 0 and 1, suitable for binary classification; Tanh (Hyperbolic Tangent), which outputs values between -1 and 1, and is zero-centered; and Softmax, which normalizes outputs for multi-class classification by converting logits to probabilities. Each function has distinct properties affecting learning and convergence in deep learning models.