AGILE - Mitigating Interference in Incremental Learning through Attention-Guided Rehearsal

The official repository for CoLLAs'24 paper. We extended the original repo DER++ with our method.

How to run?

• python main.py --seed 10 --dataset seq-cifar10 --model agile --buffer_size 200 --load_best_args
  --tensorboard --notes 'AGILE baseline'

Setup

• Use ./utils/main.py to run experiments.
• Use argument --load_best_args to use the best hyperparameters from the paper.
• New models can be added to the models/ folder.
• New datasets can be added to the datasets/ folder.

Models

• Attention-Guided Incremental Learning (AGILE)

Datasets

Class-Il / Task-IL settings

 + Sequential CIFAR-10
 + Sequential CIFAR-100
 + Sequential Tiny ImageNet