Why Bound has different results ?

[ChenJunzhi-buaa]

Is "Bound" joint-training 100 classes？Why Bound has different results ?

thank you！

[ChenJunzhi-buaa]

I use the same setting as below，but the joint training result is only 0.71 ？

[G-U-N]

Bound means: incremental learning with all the old samples reserved. Different protocols with different incremental sessions, therefore, have different average accuracies.

[ChenJunzhi-buaa]

Thank you !
So，What is your exemplars manage policy？Herding or Reservoir sampling？
What‘s more，in a batch of 128，what is the propotion of new data and old data? Or, is the batch_size only the number of new data?

[ChenJunzhi-buaa]

Bound means: incremental learning with all the old samples reserved. Different protocols with different incremental sessions, therefore, have different average accuracies.

Bound is the average acc of "incremental learning with all the old samples reserved?
And the last acc of "incremental learning with all the old samples reserved", is about 70 ?