Online Resource Allocation for Edge Intelligence with Colocated Model Retraining and Inference

Prerequisites

pytorch, numpy, thop

Usage

This is the implementation of ORRIC using the trained model in folder models.

The corresponding dataset can be found at https://zenodo.org/records/2535967. Please decompress CIFAR-10-C.tar in the root directory to obtain the CIFAR-10-C folder.

# ORRIC implementation
python train_inference_two_model.py

Others:

# teacher (resnet_50) training
python res50_teacher_training.py
# student (mobilenet_v2) training
python mobile_student_training.py

# measure the MACs
python MACs.py
# measure the time
python measure_time.py
# measure the performance
python performance.py

Slides

Citation

@INPROCEEDINGS{cai2024ORRIC,
  author={Huaiguang Cai and
          Zhi Zhou and
          Qianyi Huang},
  booktitle={IEEE INFOCOM 2024 - IEEE Conference on Computer Communications}, 
  title={Online Resource Allocation for Edge Intelligence with Colocated Model Retraining and Inference}, 
  year={2024},
  pages={1900-1909},
  doi={10.1109/INFOCOM52122.2024.10621206}}

License

This project is licensed under the MIT License - see the LICENSE file for details