In Situ Compression Artifact Removal in Scientific Data Using Deep Transfer Learning and Experience Replay
Machine Learning: Science and Technology, 2021
This paper proposes a unified in situ compression-artifact-removal framework that combines fully convolutional networks with scalable training, transfer learning, and experience replay, and shows on JPEG-compressed climate and nuclear reactor simulation data that a transfer-trained model incrementally updated during the reactor simulation substantially outperforms compressed sensing postprocessing (mean PSNR of 42.438 vs. 27.725).
BibTeX
@article{madireddy2021situ,
title={In situ compression artifact removal in scientific data using deep transfer learning and experience replay},
author={Madireddy, Sandeep and Hwan Park, Ji and Lee, Sunwoo and Balaprakash, Prasanna and Yoo, Shinjae and Liao, Wei-keng and Hauck, Cory D and Paul Laiu, M and Archibald, Richard},
journal={Machine Learning: Science and Technology},
volume={2},
number={2},
pages={025010},
year={2021},
publisher={IOP Publishing}
}