AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Neural Bootstrapper

Minsuk Shin, Hyungjoo Cho, Hyun-seok Min, Sungbin Lim

NeurIPS 2021 · Label-efficient and robust learning › Recognition and long-tail · approach: Confidence calibration objectives · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Use NeuBoots instead of amortized bootstrapping with B=5Rotated MNIST classification · Rotated MNISTtest error: 16.82 -> 14.98±0.31Improved
Increase bootstrap samples from B=1 to B=25 (ablation)NeuBoots · Rotated MNISTtest error: 17.94±0.74 -> 14.45±0.31Improved
Add calibration to NeuBoots (ablation)CIFAR-10 to SVHN OOD detection · CIFAR-10 to SVHN OODAUROC: 97.26 -> 99.14Improved
Use NeuBoots instead of DeepEnsemblePASCAL VOC 2012 semantic segmentation, ResNet-50 · PASCAL VOC 2012mIoU (%): 90.09±0.61 -> 90.14±2.17No clear effect
Use NeuBoots instead of DeepEnsemble_CENIH3T3 semantic segmentation, U-ResNet · NIH3T3mIoU (%): 65.71±1.69 -> 67.78±1.01Improved
Use NeuBoots instead of DeepEnsemble_CENIH3T3 semantic segmentation, SCNAS · NIH3T3mIoU (%): 68.66±2.58 -> 70.80±1.58Improved
Use NeuBoots instead of DeepEnsembleCIFAR-100 calibration, ResNet-110 · CIFAR-100ECE (%): 4.74±0.17 -> 8.13±0.28Hurt
Use NeuBoots instead of MCDropCIFAR-100 prediction · CIFAR-100relative prediction time: 5.0 -> 1.2Improved
Use NeuBoots instead of DeepEnsembleCIFAR-10 calibration, DenseNet-100 · CIFAR-10ECE (%): 0.54 -> 2.87Hurt
Use NeuBoots instead of MCDropCIFAR-10 test-time bootstrap outputs, ResNet-34 · CIFAR-10 test setprediction time for 100 outputs (s): 112s -> 1.9sImproved
Use NeuBoots instead of DeepEnsembleResNet-110 training · CIFAR-100training time (sec): 145.5 -> 29.2Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy