Weekly PhD meeting: Tom and Luis – 1 Dec 2016
=============================================
What I have done:
=================
I have been running experiments to validate what is the best learning rate range for the MICCAI robotic dataset. Two things are interesting about this: to experimentally discover if the fixed rate is worse than the Cyclical Learning Rate or not, what is the best range for the CLR. It is also interesting to obtain a curve where you can see the Dice coefficient rise and break, so that the rule by Bengio of using as max_lr half of the last lr that converges can be used.
Next steps:
===========
– Get frame based statistics for the validation set.
– Read ESM by Malis
Weekly meeting 1 Dec 2016