search

xyutao / fscil

Official repository for Few-Shot Class-Incremental Learning (FSCIL)
221 stars 36 forks source link

About neural gas mechanism #3

Closed petercheng85 closed 4 years ago

petercheng85 commented 4 years ago

I have the honor to read your work! And I have a simple but maybe stupid question about the mechanism of neural gas. In most CIL works, they set a parameter for the size of memory with fixed capacity. I wonder how to control the the memory size for old data in neural gas (e.g. node deletion)?

Thank you for taking the time to read this!

xyutao commented 4 years ago

For memory efficiency, the adjacent nodes with the same label can be merged to a single node by weighted aggregating, but it is not necessary for the few-shot setting. You just need to guarantee each new class has at least one corresponding NG node.

For the experimental comparison under a fixed capacity, we simply fix the memory size M = (400 + C) for all methods, where 400 is the number of base NG nodes and C is the number of all encountered new classes.

petercheng85 commented 4 years ago

Thanks for your reply!