source code

[lyc6749]

can you provide some source code for the step of getting the first model for each step size and iteration？

[fmcp]

Hi, for the first steps (without incremental learning), I used this code: https://github.com/zhanghang1989/ResNet-Matconvnet

[lyc6749]

when I use the learning rate 0.01 (80 epoch), 0.001 (80 epochs), 0.0001 (40 epochs) to train the first imdb of batch20 on cifar，I can not gain the same accuracy as you.

[fmcp]

I run 100 epochs using the learning rate commented on the paper (and included in the code for the incremental steps).

Don’t forget to include the L2 loss and random noise to the gradients.