Skip to content

Multi-level Online Sequential Experts (MOSE) for online continual learning problem. (CVPR2024)

Notifications You must be signed in to change notification settings

AnAppleCore/MOSE

Repository files navigation

MOSE

Official implementation of MOSE for online continual learning (CVPR2024).

Introduction

MOSE

Multi-level Online Sequential Experts (MOSE) cultivates the model as stacked sub-experts, integrating multi-level supervision and reverse self-distillation. Supervision signals across multiple stages facilitate appropriate convergence of the new task while gathering various strengths from experts by knowledge distillation mitigates the performance decline of old tasks.

Usage

Requirements

  • python==3.8
  • pytorch==1.12.1
pip install torch==1.10.1+cu111 torchvision==0.11.2+cu111 torchaudio==0.10.1 -f https://download.pytorch.org/whl/cu111/torch_stable.html
pip install -r requirements.txt

Training and Testing

Split CIFAR-100

python main.py \
--dataset           cifar100 \
--buffer_size       5000 \
--method            mose \
--seed              0 \
--run_nums          5 \
--gpu_id            0

Split TinyImageNet

python main.py \
--dataset           tiny_imagenet \
--buffer_size       10000 \
--method            mose \
--seed              0 \
--run_nums          5 \
--gpu_id            0

Acknowledgement

Thanks the following code bases for their framework and ideas:

Citation

If you found this code or our work useful, please cite us:

@inproceedings{yan2024orchestrate,
  title={Orchestrate Latent Expertise: Advancing Online Continual Learning with Multi-Level Supervision and Reverse Self-Distillation},
  author={Yan, Hongwei and Wang, Liyuan and Ma, Kaisheng and Zhong, Yi},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={23670--23680},
  year={2024}
}

Contact

If you have any questions or concerns, please feel free to contact us or leave an issue:

About

Multi-level Online Sequential Experts (MOSE) for online continual learning problem. (CVPR2024)

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages