Paper Notes
A notebook for some good papers I have read, including their key points and English writing.
1. Network Architecture Designed Manually
1.1 Residual
1.1.1 [Resnet-v1] Deep Residual Learning for Image Recognition
1.1.2 [Resnet-v2] Identity Mappings in Deep Residual Networks
1.2 Inception
1.2.1 [Inception-v3] Rethinking the Inception Architecture for Computer Vision
1.2.2 [Inception-v4] Inception-ResNet and the Impact of Residual Connections on Learning
1.2.3 [Xception] Deep Learning with Depthwise Separable Convolutions
1.3 Small models
1.3.1 [MobileNet-v1] Efficient Convolutional Neural Networks for Mobile Vision Applications
1.3.2 [MobileNet-v2] Inverted Residuals and Linear Bottlenecks
1.3.3 [ShuffleNet] An Extremely Efficient Convolutional Neural Network for Mobile Devices
1.4 Others
1.4.1 [Net2Net] Accelerating Learning via Knowledge Tranfer
2. Neural Architecture Search
2.1 with Reinforcement Learning
2.1.1 [NasNet-450GPUs/4days]Learning Transferable Architectures for Scalable Image Recognition
2.1.2 [1GPUs/0.5days] Efficient Neural Architecture Search via Parameters Sharing
2.2 with Evolutionary Algorithm
2.2.1 [same to 2.1.1] Regularized Evolution for Image Classifier Architecture Search
2.2.2 [2 times faster than 2.1.1] Progressive Neural Architecture Search