Welcome to CLArena (Continual Learning Arena)
An open-source machine learning package for continual learning research
CLArena (Continual Learning Arena) is a open-source Python package for Continual Learning (CL) research, solely developed by myself. In this package, I provide a integrated environment and various APIs to conduct CL experiments for research purposes, as well as implemented CL algorithms and datasets that you can give it a spin immediately. Check out what datasets and algorithms are implemented in this package:
Continual learning is a paradigm of machine learning that deals with learning new tasks sequentially without forgetting previous ones. Feel free to check out my beginner’s guide for continual learning if you haven’t know much about CL yet.
The package is powered by:
- PyTorch Lightning - a lightweight PyTorch wrapper for high-performance AI research. It removes boilerplate part of PyTorch code like batch looping, defining optimisers and losses, training strategies, so you can focus on the core algorithm part. It also keeps scalability for customisation if needed.
- Hydra - a framework for organising configurations elegantly. It can turn parameters from Python command line into hierarchical config files, which is nice for deep learning as there are usually tons of hyperparameters.
This package is adapted from the codes in my several years’ continual learning research during PhD. I have published a paper in continual learning with these codes, please check out if you are interested: Paper: AdaHAT.
Back to top