Chen Liu
Image reconstruction has long been treated as a valid and effective auxiliary task to facilitate the learning of useful and meaningful features for image classification. We are re-thinking this paradigm to see if we can use other auxiliary tasks that either emperically yields better performances on the primary task (classification), or poses different constraints on the learning process.
Please feel free to use this repository as you like to do your own experiments, try out other auxiliary tasks, test on different primary tasks, etc.
The preliminary work has been submitted to ArXiv. "Fourier Transform Approximation as an Auxiliary Task for Image Classification". If you have suggestions on what experiements/analyses to perform to further improve this manuscript, please let us know.
@article{liu2021fourier,
title={Fourier Transform Approximation as an Auxiliary Task for Image Classification},
author={Liu, Chen},
journal={arXiv e-prints},
pages={arXiv--2106},
year={2021}
}
We used anaconda environment. The step-by-step instructions are written in dependencies.txt.