⚙️ 02 — Training Neural Networks

Training means adjusting weights to reduce loss.

---

Training Loop

1. Forward pass
2. Calculate loss
3. Backpropagation
4. Optimizer updates weights
5. Repeat for many epochs

---

Key Terms

Term	Meaning
Epoch	one pass through training data
Batch	subset of data used per update
Learning rate	update step size
Optimizer	algorithm that updates weights
Loss function	objective to minimize

---

Common Losses

Task	Loss
Regression	MSE / MAE
Binary classification	binary cross-entropy
Multiclass classification	categorical cross-entropy

---

Common Problems

• learning rate too high -> unstable
• learning rate too low -> slow
• too many epochs -> overfitting
• too little data -> poor generalization

---

Next

➡️ 03-keras-quickstart