A study on the following problems: what the memorization problem is in meta-learning; why memorization problem happens; and how we can prevent it. (ICLR 2020)
Thanks first for your work! I've tried to run the pose experiments maml_bbb.py with both tf 1.14.0 and tf 2.5.0, it seems that the code is neither compatible with both versions. I'm wondering if the code is sort of outdated or how do you run them successfully? Thanks
First of all, I really appreciate that the authors present such great work. And I have a question about the implementations, does the released code include the part of meta regularization on activation? It seems only have the regularization on weights. I'm not sure if it is my misunderstanding, please kindly point me out. Any suggestions would be very appreciated.