Summarizing Stream Data for Memory-Restricted Online Continual Learning

Official implementation of "Summarizing Stream Data for Memory-Restricted Online Continual Learning"

Highlights ✨

• SSD is accepted by AAAI 2024!
• SSD summarizes the knowledge in the stream data into informative images for the replay memory.
• Through maintaining the consistency of training gradients and relationship to the past tasks, the summarized samples are more representative for the stream data compared with original images.
• SSD significantly enhances the replay effects for online continual learning methods with limited extra computational overhead.

Datasets

Online Class Incremental

• Split CIFAR100
• Split Mini-ImageNet
• Split Tiny-ImageNet

Data preparation

• CIFAR10 & CIFAR100 will be downloaded during the first run
• Mini-ImageNet: Download from https://www.kaggle.com/whitemoon/miniimagenet/download, and place it in datasets/mini_imagenet/
• Tiny-ImageNet: Download from http://cs231n.stanford.edu/tiny-imagenet-200.zip, place it in datasets/tiny-imagenet-200/. Copy process_tiny_imagenet.py to the directory and run it to suppress the dataset into pickles

Run commands

Detailed descriptions of options can be found in the SSD section in general_main.py

Sample commands to run algorithms on Split-CIFAR100

python general_main.py --data cifar100 --cl_type nc --agent SSCR --retrieve random --update summarize --mem_size 1000 --images_per_class 10 --head mlp --temp 0.07 --eps_mem_batch 100 --lr_img 4e-3 --summarize_interval 6 --queue_size 64 --mem_weight 1 --num_runs 10

Acknowledgement

This project is mainly based on online-continual-learning

Citation

If you find this work helpful, please cite:

@article{gu2023summarizing,
  title={Summarizing Stream Data for Memory-Restricted Online Continual Learning},
  author={Gu, Jianyang and Wang, Kai and Jiang, Wei and You, Yang},
  journal={arXiv preprint arXiv:2305.16645},
  year={2023}
}