MTL-Split: Multi-Task Learning for Edge Devices using Split Computing

Official implementation of the paper MTL-Split: Multi-Task Learning for Edge Devices using Split Computing accepted at the 61st Design Automation Conference (DAC 2024).

Installation

1. Repository setup:

• $ git clone https://github.com/intelligolabs/MTL-Split
• $ cd MTL-Split
• Download the 3D Shapes dataset from Google Cloud Storage. Then, add the dataset path on line 27 of the file shapes.py
• Download the MEDIC dataset from https://crisisnlp.qcri.org/medic/
• Download the FACES dataset from https://faces.mpdl.mpg.de/imeji/

2. Conda environment setup:

• $ conda create -n mtl_split python=3.10
• $ conda activate mtl_split
• $ pip install -r requirements.txt

Optionally, you can also log the training and evaluation to wandb.

• Update line 102 of the file main.py, specifying project='' and entity=''

Run MTL-Split

To run MTL-Split, use the file main.py. In particular, the launch.sh file contains two examples (STL & MTL) of a launch script example that you can use to modify the default configuration.

Authors

Luigi Capogrosso¹, Enrico Fraccaroli^1,2, Samarjit Chakraborty², Franco Fummi¹, Marco Cristani¹

¹ Department of Engineering for Innovation Medicine, University of Verona, Italy

² Department of Computer Science, The University of North Carolina at Chapel Hill, USA

¹ [email protected], ² [email protected], [email protected]

Citation

If you use MTL-Split, please, cite the following paper:

@article{capogrosso2024mtl,
  title={MTL-Split: Multi-Task Learning for Edge Devices using Split Computing},
  author={Capogrosso, Luigi and Fraccaroli, Enrico and Chakraborty, Samarjit and Fummi, Franco and Cristani, Marco},
  journal={arXiv preprint arXiv:2407.05982},
  year={2024}
}