TAMiL

Official repository for "Task-Aware Information Routing from Common Representation Space in Lifelong Learning", ICLR 2023

How to run?

• python main.py --seed 10 --dataset seq-tinyimg --model tam --buffer_size 200 --load_best_args
  --tensorboard --pretext_task mse --notes 'experiment 1'

Setup

• Use ./main.py to run experiments.
• Use argument --load_best_args to use the best hyperparameters from the paper.
• Use --evaluate to load and evaluate the model

Datasets

Class-Il / Task-IL settings

• Sequential CIFAR-10
• Sequential CIFAR-100
• Sequential Tiny-ImageNet
• Sequential Core50

Cite Our Work

If you find the code useful in your research, please consider citing our paper:

@inproceedings{
  bhat2023taskaware,
  title={Task-Aware Information Routing from Common Representation Space in Lifelong Learning},
  author={Prashant Shivaram Bhat and Bahram Zonooz and Elahe Arani},
  booktitle={The Eleventh International Conference on Learning Representations },
  year={2023},
  url={https://openreview.net/forum?id=-M0TNnyWFT5}
}