topic connectionist ★ seed

Convolutional Neural Networks

Networks using convolution operations to extract spatial features. Inspired by the visual cortex. Revolutionized computer vision from LeNet (1998) to ResNet (2015).

#cnn #convolution #computer-vision

Sub-topics

LeNet (1998) concept

Yann LeCun's 1998 CNN for handwritten digit recognition. Deployed in ATMs nationwide. Proved convolutions could learn hierarchical visual features.

AlexNet (2012) concept

Alex Krizhevsky's deep CNN won ImageNet 2012 by a massive margin (top-5 error 15.3% vs 26.2%). Used GPUs, ReLU, and dropout. Ignited the deep learning revolution.

ResNet (2015) concept

Kaiming He's residual networks introduced skip connections, enabling training of 152-layer networks. Achieved 3.57% top-5 error, surpassing human performance on ImageNet.