NeurIPS2025-LEAR

The official implementation for "Learning Expandable and Adaptable Representations for Continual Learning" (NeurIPS2025)

---

▶️ Usage

1. Create env and install requirements

conda create -n LEAR python=3.10
conda activate LEAR
pip install -r requirements.txt

2. Run the example training script

bash LEAR.sh

Project structure overview

LEAR/
├── backbone/                 # Pre-trained backbone models
│   ├── LEAR.py               # LEAR backbone implementation
│   └── ...
├── datasets/                 # Dataset loaders
|   ├── init.py               # Modify domain sequence                
│   └── ...
├── models/                   # CL Method implementations
│   └── LEAR.py               # LEAR method implementation
├── utils/                    # Helper tools
|   ├── train_domain.py       # Training scripts                
│   └── ...
├── main_domain.py            # Main entry
├── LEAR.sh
└── README.md

---

📝 Citation

If you find this repository helpful, please click the ⭐Star and cite our paper:

@inproceedings{yulearning,
  title={Learning Expandable and Adaptable Representations for Continual Learning},
  author={Yu, Ruilong and Liu, Mingyan and Ye, Fei and Bors, Adrian G and Hu, Rongyao and others},
  booktitle={The Thirty-ninth Annual Conference on Neural Information Processing Systems}
}

---

🙏 Acknowledgement

Thanks for the awesome continual learning framework Mammoth.