Sparse Coding with Gated Learned ISTA
Abstract
In this paper, we study the learned iterative shrinkage thresholding algorithm (LISTA) for solving sparse coding problems. Following assumptions made by prior works, we first discover that the code components in its estimations may be lower than expected, i.e., require gains, and to address this problem, a gated mechanism amenable to theoretical analysis is then introduced. Specific design of the gates is inspired by convergence analyses of the mechanism and hence its effectiveness can be formally guaranteed. In addition to the gain gates, we further introduce overshoot gates for compensating insufficient step size in LISTA. Extensive empirical results confirm our theoretical findings and verify the effectiveness of our method.
Cite
Text
Wu et al. "Sparse Coding with Gated Learned ISTA." International Conference on Learning Representations, 2020.Markdown
[Wu et al. "Sparse Coding with Gated Learned ISTA." International Conference on Learning Representations, 2020.](https://mlanthology.org/iclr/2020/wu2020iclr-sparse/)BibTeX
@inproceedings{wu2020iclr-sparse,
title = {{Sparse Coding with Gated Learned ISTA}},
author = {Wu, Kailun and Guo, Yiwen and Li, Ziang and Zhang, Changshui},
booktitle = {International Conference on Learning Representations},
year = {2020},
url = {https://mlanthology.org/iclr/2020/wu2020iclr-sparse/}
}