Neural networks

Understands a neuron as a weighted sum + activation; knows that networks learn by adjusting weights.

Implements a 2-layer MLP from scratch in NumPy; understands sigmoid, ReLU, softmax activations.

Builds and trains networks with PyTorch/Keras; applies batch norm, dropout, weight init; debugs gradient issues.

Designs custom layers, loss functions, training loops; applies advanced init schemes (Kaiming/Xavier); understands neural tangent kernel.

Contributes to neural architecture search, mechanistic interpretability, or theoretical deep learning.