Exercises: Deep Learning

Concept Drills

• Draw a neural network and label weights, bias, activation, and output.
• Explain forward propagation.
• Explain backpropagation without equations first, then with equations.
• Compare SGD, Adam, and learning rate schedules.

Coding Drills

• Build a two-layer neural network from scratch.
• Train a classifier with a deep learning framework.
• Add dropout and compare results.
• Add batch normalization and compare results.
• Export a trained model and load it for inference.

Failure Drills

• Use a learning rate that is too high.
• Use a learning rate that is too low.
• Train without normalization.
• Overfit a tiny dataset intentionally.

Output

• Experiment tracking table.
• Loss curve plots.
• Inference API demo.