Open ifeherva opened 4 years ago
I noticed that on mnist most optimizers do not reach 100% train accuracy (or oscillate) while in my implementation I get 100% train accuracy after like 10 epochs.
Is it possible to publish the code/setup for the mnist experiments as well?
I noticed that on mnist most optimizers do not reach 100% train accuracy (or oscillate) while in my implementation I get 100% train accuracy after like 10 epochs.
Is it possible to publish the code/setup for the mnist experiments as well?