gmumcl
Research software · Continual learning
gmumcl is a lightweight continual-learning framework for PyTorch, developed at GMUM. It provides five incremental learning scenarios (class, task, domain, instance, and permuted), an accuracy-matrix trainer, a compact plugin API for methods, and reference implementations of standard continual-learning baselines such as EWC, SI, MAS, LwF, and GPM. It builds on ActivationScope for activation-level bookkeeping.
ActivationScope
Research software · Neural network analysis
ActivationScope is a high-performance PyTorch plugin library for tracking, storing, and analyzing intermediate activations. It is designed for efficient activation analysis on large neural networks, with a Python/C++ implementation and online reduction functionality that keeps memory use bounded during training or inference.