Dropout as a Bayesian Approximation: Representing Model Uncertainty in Deep Learning
Yarin Gal, Zoubin Ghahramani · ICML 2016 · 2016
- Status
- To read
- Topics
- uncertainty
- Feeds
- P2 · NeurIPS 2027
What they did
Shows that dropout training approximates Bayesian inference in deep Gaussian processes, so keeping dropout on at test time (Monte Carlo dropout) gives model uncertainty at little extra cost.
My note
No note yet.
Code
- Repository
- yaringal/DropoutUncertaintyExps
- Framework
- —
- Licence
- See the repository
- My status
- Not tried
- Note
- Experiment code from the paper.
BibTeX
@inproceedings{gal2016-mc-dropout,
title = {{Dropout as a Bayesian Approximation: Representing Model Uncertainty in Deep Learning}},
author = {Yarin Gal and Zoubin Ghahramani},
booktitle = {ICML 2016},
year = {2016},
eprint = {1506.02142},
archivePrefix = {arXiv}
}