Index page

GoogLe Net

[3] GoogLe Net Szegedy, Christian, et al. "Going deeper with convolutions." Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2015. Paper

Notes

• Inception module
• Network in Network -> 1x1 convolutions
• Solve overfitting+large mem -> replace FC layers with sparse
• Ratio of 3x3 and 5x5 convolutions should increase in higher layers
• Dimensionality reduction
• Merge outputs of best models
• Approximate expected optimal sparse structure by using dense blocks

• Shallow and narrower Architecture

• Architecture