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

It is now well known that neural networks can be wrong with high confidencein their predictions, leading to poor calibration. The most common post-hocapproach to compensate for this is to perform temperature scaling, whichadjusts the confidences of the predictions on any input by scaling the logitsby a fixed value. Whilst this approach typically improves the averagecalibration across the whole test dataset, this improvement typically reducesthe individual confidences of the predictions irrespective of whether theclassification of a given input is correct or incorrect. With this insight, webase our method on the observation that different samples contribute to thecalibration error by varying amounts, with some needing to increase theirconfidence and others needing to decrease it. Therefore, for each input, wepropose to predict a different temperature value, allowing us to adjust themismatch between confidence and accuracy at a finer granularity. Furthermore,we observe improved results on OOD detection and can also extract a notion ofhardness for the data-points. Our method is applied post-hoc, consequentlyusing very little computation time and with a negligible memory footprint andis applied to off-the-shelf pre-trained classifiers. We test our method on theResNet50 and WideResNet28-10 architectures using the CIFAR10/100 andTiny-ImageNet datasets, showing that producing per-data-point temperatures isbeneficial also for the expected calibration error across the whole test set.Code is available at: https://github.com/thwjoy/adats.

Sample-dependent Adaptive Temperature Scaling for Improved Calibration