PaperGym
← Back to Foundations
Module 3 of 6

Activation Functions

Implement the nonlinearities that make neural networks work.

Progress0/3 drills · 0%
FREE~18 min·Beginner

What you'll learn

  • ReLU and its variants (Leaky ReLU, GELU)
  • Sigmoid and tanh
  • Softmax with numerical stability

Drills

1

ReLU & Variants

Implement relu, leaky_relu, and gelu — the most common activation functions in modern networks.

Easy🕐 6m10 pts
2

Sigmoid & Tanh

Implement sigmoid and tanh from scratch and verify the identity tanh(x) = 2*sigmoid(2x) - 1.

Easy🕐 6m10 pts
3

Softmax

Implement numerically stable softmax for 1D and 2D arrays — the output layer of every classifier.

Medium🕐 8m15 pts

📚 Resources