search

GMvandeVen / continual-learning

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.
MIT License
1.54k stars 310 forks source link

Single head or multihead task incremental #27

Closed COD1995 closed 11 months ago

COD1995 commented 11 months ago

Hi, there

Is this experiment on task incremental single head or multi head?

Thanks

GMvandeVen commented 11 months ago

Hi, in the accompanying article, most experiments in the task-incremental learning scenario use a multi-headed output layer. Exception are experiments in Supplemantary Note 7, in which there are experiments in the task-incremental learning scenario that use a single-headed output layer. In the code, when you select the task-incremental learning scenario (using --scenario=task), by default a multi-headed output layer will be used, but in this scenario you can instead use a single-headed output layer by additionally using the option --singlehead. Note that experiments in the domain- or class-incremental learning scenario always use a single-headed output layer. Hope this helps!