Exploiting Task Relationships for Continual Learning Using Transferability-Aware Task Embeddings

Environment preparation

conda install pytorch torchvision torchaudio pytorch-cuda=12.1 -c pytorch -c nvidia
pip install hydra-core lightning tqdm matplotlib tensorboard
pip install tensorboardX
pip install scikit-learn numpy

Run experiments

python run.py learning.epoch=60 dataset.num_tasks=10 dataset.num_classes_per_task=20 seed=42

Here the config settings can be changed as you want. All configs including data path could also be set using the config files in ./conf/.