“Hey Alex, do you know what AlexNet is?” The little spirit asked Alex.
“AlexNet is a game changer. Many years ago, everyone were using basic machine learning models to recognize images β and they were struggling. Then comes AlexNet, saying, βLet deep learning handle this!β And it did. Spectacularly!”
“AlexNet is a Convolutional Neural Network (CNN) designed for image classification.But it wasn’t invented by our uncle Alex.” Lisa said, to which his brother wonder, “So, who invented it?”
“It was developed by Alex Krizhevsky, Ilya Sutskever, and Geoffrey Hinton, and it won the ImageNet Challenge in 2012 by a landslide β slashing the error rate from 26.2% to 15.3%.” Lisa replied, why they still working in the magic garden.
“So, it was the game changer because it was the first?” Luis asked, to which Lisa replied. “There are some are remarkable things about it as well, in addition to the fact that AlexNet also illutrated that deep architecture could work on large datasets.”
“It used GPUs for training β way faster than CPUs. Also, using ReLU activation function instead of tanh/sigmoid also speedup training”
“5 Convolutional Layers help extract features like edges, textures, and shapes. Max Pooling downsamples the image to focus on key features. ReLU Activation: Adds non-linearity and speeds up training!” Alex said.
“Yes, and Dropout + Data Augmentation helped reduce overfitting! 3 Fully Connected Layers and Softmax Output help make final decisions about what the image is.” Lisa added.
Think of AlexNet as a layered cake with 8 delicious layers:
- π° 5 Convolutional Layers: Extract features like edges, textures, and shapes
- π§ Max Pooling: Downsamples the image to focus on key features
- π₯ ReLU Activation: Adds non-linearity and speeds up training
- π§’ Dropout: Prevents overfitting by randomly turning off neurons
- π§ 3 Fully Connected Layers: Makes final decisions about what the image is
- π― Softmax Output: Chooses from 1,000 possible categories