📌 1. AlexNet

• AlexNet was the first deep CNN to achieve breakthrough results on ImageNet.
• It used 8 layers (5 convolutional, 3 fully connected).
• First to combine GPUs for training, and introduced data augmentation and dropout for regularization.

✅ Impact: Made deep CNNs practical for large datasets.

🌀 2. Pooling Layer

• Pooling layers reduce spatial dimensions (height & width) while keeping depth.
• Example: Max pooling or Average pooling.
• No learnable parameters because they only apply a fixed operation (max or average) — no weights to train! and because pooling layers has no weights.

✅ Helps with translation invariance and reduces overfitting.

⚡ 3. First Use of ReLU

• ReLU (Rectified Linear Unit) was popularized by AlexNet.
• Faster training and mitigated the vanishing gradient problem that plagued sigmoid/tanh activations.

🔍 4. ZFNet

• ZFNet (Zeiler & Fergus 2013) fine-tuned AlexNet by:
  • Improving hyperparameters like filter size & stride.
  • Visualizing intermediate feature maps to better understand learned filters.

✅ Showed the value of interpretability tools in CNNs.