Calibrating deep neural networks using focal loss
Miscalibration -- a mismatch between a model's confidence and its correctness -- of Deep Neural Networks (DNNs) makes their predictions hard to rely on. Ideally, we want networks to be accurate, calibrated and confident. We show that, as opposed to the standard cross-entropy loss, focal loss (L...
| Main Authors: | , , , , , |
|---|---|
| Format: | Conference item |
| Language: | English |
| Published: |
Curran Associates
2020
|
| _version_ | 1797061568918192128 |
|---|---|
| author | Mukhoti, J Kulharia, V Sanyal, A Golodetz, S Torr, PHS Dokania, PK |
| author_facet | Mukhoti, J Kulharia, V Sanyal, A Golodetz, S Torr, PHS Dokania, PK |
| author_sort | Mukhoti, J |
| collection | OXFORD |
| description | Miscalibration -- a mismatch between a model's confidence and its correctness -- of Deep Neural Networks (DNNs) makes their predictions hard to rely on. Ideally, we want networks to be accurate, calibrated and confident. We show that, as opposed to the standard cross-entropy loss, focal loss (Lin et al., 2017) allows us to learn models that are already very well calibrated. When combined with temperature scaling, whilst preserving accuracy, it yields state-of-the-art calibrated models. We provide a thorough analysis of the factors causing miscalibration, and use the insights we glean from this to justify the empirically excellent performance of focal loss. To facilitate the use of focal loss in practice, we also provide a principled approach to automatically select the hyperparameter involved in the loss function. We perform extensive experiments on a variety of computer vision and NLP datasets, and with a wide variety of network architectures, and show that our approach achieves state-of-the-art calibration without compromising on accuracy in almost all cases. Code is available at https://github.com/torrvision/focal_calibration. |
| first_indexed | 2024-03-06T20:33:04Z |
| format | Conference item |
| id | oxford-uuid:31b60972-2e7f-4388-8551-1f3b59020245 |
| institution | University of Oxford |
| language | English |
| last_indexed | 2024-03-06T20:33:04Z |
| publishDate | 2020 |
| publisher | Curran Associates |
| record_format | dspace |
| spelling | oxford-uuid:31b60972-2e7f-4388-8551-1f3b590202452022-03-26T13:09:41ZCalibrating deep neural networks using focal lossConference itemhttp://purl.org/coar/resource_type/c_5794uuid:31b60972-2e7f-4388-8551-1f3b59020245EnglishSymplectic ElementsCurran Associates2020Mukhoti, JKulharia, VSanyal, AGolodetz, STorr, PHSDokania, PKMiscalibration -- a mismatch between a model's confidence and its correctness -- of Deep Neural Networks (DNNs) makes their predictions hard to rely on. Ideally, we want networks to be accurate, calibrated and confident. We show that, as opposed to the standard cross-entropy loss, focal loss (Lin et al., 2017) allows us to learn models that are already very well calibrated. When combined with temperature scaling, whilst preserving accuracy, it yields state-of-the-art calibrated models. We provide a thorough analysis of the factors causing miscalibration, and use the insights we glean from this to justify the empirically excellent performance of focal loss. To facilitate the use of focal loss in practice, we also provide a principled approach to automatically select the hyperparameter involved in the loss function. We perform extensive experiments on a variety of computer vision and NLP datasets, and with a wide variety of network architectures, and show that our approach achieves state-of-the-art calibration without compromising on accuracy in almost all cases. Code is available at https://github.com/torrvision/focal_calibration. |
| spellingShingle | Mukhoti, J Kulharia, V Sanyal, A Golodetz, S Torr, PHS Dokania, PK Calibrating deep neural networks using focal loss |
| title | Calibrating deep neural networks using focal loss |
| title_full | Calibrating deep neural networks using focal loss |
| title_fullStr | Calibrating deep neural networks using focal loss |
| title_full_unstemmed | Calibrating deep neural networks using focal loss |
| title_short | Calibrating deep neural networks using focal loss |
| title_sort | calibrating deep neural networks using focal loss |
| work_keys_str_mv | AT mukhotij calibratingdeepneuralnetworksusingfocalloss AT kulhariav calibratingdeepneuralnetworksusingfocalloss AT sanyala calibratingdeepneuralnetworksusingfocalloss AT golodetzs calibratingdeepneuralnetworksusingfocalloss AT torrphs calibratingdeepneuralnetworksusingfocalloss AT dokaniapk calibratingdeepneuralnetworksusingfocalloss |