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)".
This repository implements the following models:
- AdaTT [Paper]
- MMoE [Paper]
- Multi-level MMoE (an extension of MMoE)
- PLE [Paper]
- Cross-stitch [Paper]
- Shared-bottom [Paper]
To facilitate the integration and selection of these models, we have implemented a class called CentralTaskArch.
AdaTT is MIT-licensed.
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}
}