Explain the architecture and training process of deep neural networks.

Deep neural networks learn hierarchical representations through multiple layers of interconnected neurons with nonlinear activation functions.\n\n• Architecture: Input layer, hidden layers with neurons, output layer\n• Forward pass: Weighted sums, activation functions (ReLU, sigmoid, tanh)\n• Backpropagation: Compute gradients using chain rule, update weights\n• Regularization: Dropout, batch normalization, weight decay\n\nExample: For image classification, use CNN with conv layers (feature extraction), pooling (dimensionality reduction), fully connected layers (classification). Apply dropout (0.5), batch normalization, and data augmentation to prevent overfitting.