PyTorch implementation of various methods for continual learning (XdG, EWC, SI, LwF, FROMP, DGR, BI-R, ER, A-GEM, iCaRL, Generative Classifier) in three different scenarios.
Hey, Haven't found any contact of yours on your page, so asking here.
Note: This is not an issue, but a concept question?
I have a text data stream coming daily on different domains like agriculture, space, biology, etc...
I train them to detect domains, but new domains are incoming too, so how can train new data preserving the learnings of previous ones?
You have specified various forgetting prevention techniques on MNIST, which do you think would work on text data using let us say bert as an encoder?
Hey, Haven't found any contact of yours on your page, so asking here. Note: This is not an issue, but a concept question?
I have a text data stream coming daily on different domains like agriculture, space, biology, etc... I train them to detect domains, but new domains are incoming too, so how can train new data preserving the learnings of previous ones?
You have specified various forgetting prevention techniques on MNIST, which do you think would work on text data using let us say bert as an encoder?
Waiting for your reply! Thanks!