Deep Learning Essentials

Convolution → ReLU → Pooling → Flatten → Dense — full pipeline

3x3 kernel on 7x7 image — worked calculation, what features get detected

stride=1 vs stride=2 output size math, same vs valid padding

y=max(0,x), why negatives become zero, vanishing gradient fix

What each preserves, 2x2 pooling worked example, hyperparameter effect

Rescaling, tilting, zoom, blur, noise — why needed, SK Image demo

Pitch shift, noise injection, silence removal, why for GenAI

Pre-trained models as starting point, feature extraction vs fine-tuning

Regularization techniques, where to apply, effect on training

Building CNN in Keras: layers, compile, fit, evaluate, predict