continual-learning

Why we included this project

This codebase bundles implementations of several established continual learning methods, from EWC and SI to LwF, A-GEM, iCaRL, and generative replay, all in one place. That makes it a convenient base for comparing approaches without building each one from scratch. The experiments follow the three canonical incremental learning scenarios (task-, domain-, and class-incremental), which line up with the distinctions in the accompanying Nature Machine Intelligence paper. Since this is the code behind that publication and a NeurIPS tutorial, the implementations are careful and reproducible, with explicit run instructions in the README. If you are studying or teaching continual learning, this is a dependable reference to read and extend.