GMvandeVen/continual-learning

Implements task-incremental, domain-incremental, and class-incremental learning scenarios with support for both strict academic settings (non-overlapping task boundaries) and flexible task-free learning with gradual context transitions. Built on PyTorch with modular architecture enabling method combinations and custom approaches by mixing regularization (EWC, SI), replay (generative/buffer-based), and architectural components (gating, separate networks). Includes real-time training visualization via Visdom and reproducible benchmark experiments on Split/Permuted MNIST and CIFAR datasets.

1,832 stars.