https://github.com/facebookresearch/adatt
pytorch open-source library for the paper "AdaTT Adaptive Task-to-Task Fusion Network for Multitask Learning in Recommendations"
https://github.com/facebookresearch/adatt
moe mtl multitask-learning
Last synced: 6 months ago
JSON representation
pytorch open-source library for the paper "AdaTT Adaptive Task-to-Task Fusion Network for Multitask Learning in Recommendations"
- Host: GitHub
- URL: https://github.com/facebookresearch/adatt
- Owner: facebookresearch
- License: mit
- Created: 2023-06-03T00:01:08.000Z (over 2 years ago)
- Default Branch: main
- Last Pushed: 2024-08-09T04:17:17.000Z (about 1 year ago)
- Last Synced: 2025-03-23T16:01:34.629Z (7 months ago)
- Topics: moe, mtl, multitask-learning
- Language: Python
- Homepage:
- Size: 18.6 KB
- Stars: 46
- Watchers: 1
- Forks: 8
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- Contributing: CONTRIBUTING.md
- License: LICENSE
- Code of conduct: CODE_OF_CONDUCT.md
Awesome Lists containing this project
README
# AdaTT
Welcome to the AdaTT repository! This repository provides a PyTorch library for multitask learning, specifically focused on the models evaluated in the paper ["AdaTT: Adaptive Task-to-Task Fusion Network for Multitask Learning in Recommendations" (KDD'23)"](https://doi.org/10.1145/3580305.3599769).
[**[arXiv]**](https://arxiv.org/abs/2304.04959) [**[slides]**](https://drive.google.com/file/d/1I8XpxPxwhP9KXuztEguYkuMM10kiJDS7/view?usp=sharing)
## Models
This repository implements the following models:
+ AdaTT [[Paper]](https://doi.org/10.1145/3580305.3599769)
+ MMoE [[Paper]](https://dl.acm.org/doi/10.1145/3219819.3220007)
+ Multi-level MMoE (an extension of MMoE)
+ PLE [[Paper]](https://doi.org/10.1145/3383313.3412236)
+ Cross-stitch [[Paper]](https://openaccess.thecvf.com/content_cvpr_2016/papers/Misra_Cross-Stitch_Networks_for_CVPR_2016_paper.pdf)
+ Shared-bottom [[Paper]](https://link.springer.com/article/10.1023/a:1007379606734)
To facilitate the integration and selection of these models, we have implemented a class called `CentralTaskArch`.
## License
AdaTT is MIT-licensed.
## Citation
If you find AdaTT's paper or code helpful, please consider citing:
```
@article{li2023adatt,
title={AdaTT: Adaptive Task-to-Task Fusion Network for Multitask Learning in Recommendations},
author={Li, Danwei and Zhang, Zhengyu and Yuan, Siyang and Gao, Mingze and Zhang, Weilin and Yang, Chaofei and Liu, Xi and Yang, Jiyan},
journal={arXiv preprint arXiv:2304.04959},
year={2023}
}
```