Victor Schmidt & Simon Verret
Mila
3IT - Université de Sherbrooke
14 Novembre 2019
MLP
Loss
Backprop
Convolutions
Réseaux convolutionnels
GANs
Classification
Semantic Segmentation
Detection
Instance Segmentation
Image Generation
Image to Image Translation
Stride = 2
Padding = 1
Padding = same
Dilated
Transposed
Convolution arithmetic tutorial
LeCun, 1998
Allows the training of deeper networks
Less sensitive to initialization
Larger learning rates can be used
Faster and more stable convergence
Ioffe, 2015
1000 classes
~10 000 images per classes
1.4M in total
Top-1 ou Top-5 accuracy
First deep networks to win "ImageNet competition"
Krizhevsky, 2012
Even deeper, more regular kernels (3x3)
Simonyan, 2014
"We need to go deeper"
Szegedy, 2014
He, 2015
Xie, 11 Nov 2019
Transfer learning from pre-trained models
Goodfellow, 2014
Mirza, 2014
Zhu, 2017
schmidtv@mila.quebec
htttps://vict0rs.ch
More reading material
By Victor Smt
