English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Mixup~\cite{zhang2017mixup} is a recently proposed method for training deepneural networks where additional samples are generated during training byconvexly combining random pairs of images and their associated labels. Whilesimple to implement, it has been shown to be a surprisingly effective method ofdata augmentation for image classification: DNNs trained with mixup shownoticeable gains in classification performance on a number of imageclassification benchmarks. In this work, we discuss a hitherto untouched aspectof mixup training -- the calibration and predictive uncertainty of modelstrained with mixup. We find that DNNs trained with mixup are significantlybetter calibrated -- i.e., the predicted softmax scores are much betterindicators of the actual likelihood of a correct prediction -- than DNNstrained in the regular fashion. We conduct experiments on a number of imageclassification architectures and datasets -- including large-scale datasetslike ImageNet -- and find this to be the case. Additionally, we find thatmerely mixing features does not result in the same calibration benefit and thatthe label smoothing in mixup training plays a significant role in improvingcalibration. Finally, we also observe that mixup-trained DNNs are less prone toover-confident predictions on out-of-distribution and random-noise data. Weconclude that the typical overconfidence seen in neural networks, even onin-distribution data is likely a consequence of training with hard labels,suggesting that mixup be employed for classification tasks where predictiveuncertainty is a significant concern.

On Mixup Training: Improved Calibration and Predictive Uncertainty for Deep Neural Networks