Using mixup as a regularizer can surprisingly improve accuracy and out-of-distribution robustness

Using mixup as a regularizer can surprisingly improve accuracy and out-of-distribution robustness

We show that the effectiveness of the well celebrated Mixup can be further improved if instead of using it as the sole learning objective, it is utilized as an additional regularizer to the standard cross-entropy loss. This simple change not only improves accuracy but also significantly improves the...

Disgrifiad llawn

Manylion Llyfryddiaeth
Prif Awduron:	Pinto, F, Yang, H, Lim, SN, Torr, PHS, Dokania, PK
Fformat:	Conference item
Iaith:	English
Cyhoeddwyd:	Curran Associates, Inc 2023

Eitemau Tebyg

Mix-MaxEnt: improving accuracy and uncertainty estimates of deterministic neural networks
gan: Pinto, F, et al.
Cyhoeddwyd: (2021)

Placing objects in context via inpainting for out-of-distribution segmentation
gan: De Jorge, P, et al.
Cyhoeddwyd: (2024)

An impartial take to the CNN vs transformer robustness contest
gan: Pinto, F, et al.
Cyhoeddwyd: (2022)

Raising the bar on the evaluation of out-of-distribution detection
gan: Mukhoti, J, et al.
Cyhoeddwyd: (2023)

Are vision transformers always more robust than convolutional neural networks?
gan: Pinto, F, et al.
Cyhoeddwyd: (2021)

Sample-dependent adaptive temperature scaling for improved calibration
gan: Joy, T, et al.
Cyhoeddwyd: (2023)

Deep facial phenotyping with mixup augmentation
gan: Campbell, J, et al.
Cyhoeddwyd: (2023)

Data augmentation and mixup techniques for stock returns prediction
gan: Tan, Regan Yik Rong
Cyhoeddwyd: (2024)

Computation-efficient knowledge distillation via uncertainty-aware mixup
gan: Xu, Guodong, et al.
Cyhoeddwyd: (2023)

GDumb: A simple approach that questions our progress in continual learning
gan: Prabhu, A, et al.
Cyhoeddwyd: (2020)

Discovering class-specific pixels for weakly-supervised semantic segmentation
gan: Chaudhry, A, et al.
Cyhoeddwyd: (2017)

Semantic-discriminative mixup for generalizable sensor-based cross-domain activity recognition
gan: Lu, Wang, et al.
Cyhoeddwyd: (2023)

A surprisingly robust trick for the Winograd Schema Challenge
gan: Kocijan, V, et al.
Cyhoeddwyd: (2019)

Continual learning in low-rank orthogonal subspaces
gan: Chaudhry, A, et al.
Cyhoeddwyd: (2020)

Surprise
gan: Ross, Wendy, et al.
Cyhoeddwyd: (2022)

Surprises.
gan: Weiskrantz, L
Cyhoeddwyd: (2008)

No Surprises
gan: Wells, Ian
Cyhoeddwyd: (2021)

Mirror Descent view for Neural Network quantization
gan: Ajanthan, T, et al.
Cyhoeddwyd: (2021)

How robust is unsupervised representation learning to distribution shift?
gan: Shi, Y, et al.
Cyhoeddwyd: (2023)

Regularization Through Feature Knock Out
gan: Wolf, Lior, et al.
Cyhoeddwyd: (2005)

On surprise-hacking
gan: Felin, T, et al.
Cyhoeddwyd: (2019)

Integral sliding mode control for improved robustness and accuracy of induction motors
gan: Mohamed, H.A.F., et al.
Cyhoeddwyd: (2008)

AirMixML: Over-the-Air Data Mixup for Inherently Privacy-Preserving Edge Machine Learning
gan: Koda, Yusuke, et al.
Cyhoeddwyd: (2022)

The Surprising Breadth of Harbingers of Failure
gan: Simester, Duncan, et al.
Cyhoeddwyd: (2021)

Robust parameterization and computation of the trifocal tensor
gan: Torr, PHS, et al.
Cyhoeddwyd: (1998)

STEER: simple temporal regularization for neural ODEs
gan: Ghosh, A, et al.
Cyhoeddwyd: (2020)

On using focal loss for neural network calibration
gan: Mukhoti, J, et al.
Cyhoeddwyd: (2020)

Calibrating deep neural networks using focal loss
gan: Mukhoti, J, et al.
Cyhoeddwyd: (2020)

RanDumb: a simple approach that questions the efficacy of continual representation learning
gan: Prabhu, A, et al.
Cyhoeddwyd: (2025)

RanDumb: random representations outperform online continually learned representations
gan: Prabhu, A, et al.
Cyhoeddwyd: (2025)

Supervised contrastive pretrained ResNet with MixUp to enhance respiratory sound classification on imbalanced and limited dataset
gan: Hu, Jinhai, et al.
Cyhoeddwyd: (2024)

Surprise, surprise! (again) The 2017 British general election exit poll
gan: Curtice, J, et al.
Cyhoeddwyd: (2017)

Jackie surprises himself
gan: Singh, Devinder
Cyhoeddwyd: (2016)

The surprise theory of everything
gan: Vedral, V
Cyhoeddwyd: (2012)

Aristophanes and the poetics of surprise
gan: Kanellakis, D
Cyhoeddwyd: (2019)

Surprises in theoretical physics/
gan: 406046 Peierls, Rudolf
Cyhoeddwyd: (1979)

Surprises in the Consumption Function.
gan: Muellbauer, J
Cyhoeddwyd: (1983)

Universal regularization methods: varying the power, the smoothness and the accuracy
gan: Cartis, C, et al.
Cyhoeddwyd: (2019)

Regularity and robustness in monotone Bayesian games
gan: Beggs, A
Cyhoeddwyd: (2015)

Regularity and robustness in monotone Bayesian games
gan: Beggs, A
Cyhoeddwyd: (2015)