ImageNet Classification with Deep Convolutional Neural Networks (AlexNet)¶
Authors: Alex Krizhevsky, Ilya Sutskever, Geoffrey E. Hinton Year: 2012 Citations: 80,000+
Summary¶
Deep 8-layer CNN winning ImageNet 2012 with dramatic accuracy jump. Introduces ReLU, dropout, GPU training, data augmentation. Triggered deep learning revolution.
Key Concepts¶
- Deep Architecture: 8 convolutional layers much deeper than prior work
- ReLU Activation: Non-saturation enabling faster training
- Dropout Regularization: Random unit dropping prevents overfitting
- GPU Training: Accelerated computation on commodity hardware
- Data Augmentation: Random crops and flips increase diversity
- End-to-End Learning: Backpropagation through all layers
Impact¶
- 80,000+ citations
- Triggered deep learning revolution
- Showed deep networks work for vision
- Introduced key techniques (ReLU, dropout)
- Drove GPU adoption
- Inspired ResNet, VGGNet, Inception
- Transformed vision from features to learned representations
Related Papers¶
- LeNet (LeCun et al., 1998)
- VGGNet (Simonyan & Zisserman, 2014)
- ResNet (He et al., 2015)