AllWeights
This is a public repository for different weight assignment schemes in multi-task learning systems. Using this library users can create objects of different weight assignment schemes and use it directly for the loss value.
Requirements
PyTorch > 1.9 Python > 3.7
Installation
Clone this library to your project using the following command
https://github.com/aminul-huq/AllWeights.git
Weight assignment techniques
| Method name | Description |
|---|---|
| equal.py | Each loss will have equal weights |
| random.py | Each loss will have random weights |
| dwa.py | End-to-End Multi-Task Learning with Attention [CVPR 2019] |
| uncertainty.py | Multi-Task Learning Using Uncertainty to Weigh Losses for Scene Geometry and Semantics [CVPR 2018] |
| reu.py | Auxiliary tasks in multi-task learning. arXiv preprint arXiv:1805.06334, 2018. |
N.B. DWA method needs to be updated.
Example
from AllWeights import *
w = Equal(3)
Loss = w([Task1_loss, Task2_loss, Task3_loss])