Packed-Ensembles for Efficient Uncertainty Estimation
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2022
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866918145905655808 |
|---|---|
| author | Laurent, Olivier Lafage, Adrien Tartaglione, Enzo Daniel, Geoffrey Martinez, Jean-Marc Bursuc, Andrei Franchi, Gianni |
| author_facet | Laurent, Olivier Lafage, Adrien Tartaglione, Enzo Daniel, Geoffrey Martinez, Jean-Marc Bursuc, Andrei Franchi, Gianni |
| contents | Deep Ensembles (DE) are a prominent approach for achieving excellent performance on key metrics such as accuracy, calibration, uncertainty estimation, and out-of-distribution detection. However, hardware limitations of real-world systems constrain to smaller ensembles and lower-capacity networks, significantly deteriorating their performance and properties. We introduce Packed-Ensembles (PE), a strategy to design and train lightweight structured ensembles by carefully modulating the dimension of their encoding space. We leverage grouped convolutions to parallelize the ensemble into a single shared backbone and forward pass to improve training and inference speeds. PE is designed to operate within the memory limits of a standard neural network. Our extensive research indicates that PE accurately preserves the properties of DE, such as diversity, and performs equally well in terms of accuracy, calibration, out-of-distribution detection, and robustness to distribution shift. We make our code available at https://github.com/ENSTA-U2IS/torch-uncertainty. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2210_09184 |
| institution | arXiv |
| publishDate | 2022 |
| record_format | arxiv |
| spellingShingle | Packed-Ensembles for Efficient Uncertainty Estimation Laurent, Olivier Lafage, Adrien Tartaglione, Enzo Daniel, Geoffrey Martinez, Jean-Marc Bursuc, Andrei Franchi, Gianni Machine Learning Deep Ensembles (DE) are a prominent approach for achieving excellent performance on key metrics such as accuracy, calibration, uncertainty estimation, and out-of-distribution detection. However, hardware limitations of real-world systems constrain to smaller ensembles and lower-capacity networks, significantly deteriorating their performance and properties. We introduce Packed-Ensembles (PE), a strategy to design and train lightweight structured ensembles by carefully modulating the dimension of their encoding space. We leverage grouped convolutions to parallelize the ensemble into a single shared backbone and forward pass to improve training and inference speeds. PE is designed to operate within the memory limits of a standard neural network. Our extensive research indicates that PE accurately preserves the properties of DE, such as diversity, and performs equally well in terms of accuracy, calibration, out-of-distribution detection, and robustness to distribution shift. We make our code available at https://github.com/ENSTA-U2IS/torch-uncertainty. |
| title | Packed-Ensembles for Efficient Uncertainty Estimation |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2210.09184 |