AI Paper Atlas

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

Backward-Compatible Prediction Updates: A Probabilistic Approach

Frederik Träuble, Julius Von Kügelgen, Matthäus Kleindessner, Francesco Locatello, Bernhard Schölkopf, Peter Vincent Gehler

NeurIPS 2021 · Label-efficient and robust learning › Recognition and long-tail · approach: Confidence calibration objectives · 10 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 CR 10 update rule instead of Replaceposterior-based updates · ImageNet->ImageNet, budget 100%cumulative negative flips: 24214 -> 2112Improved
use CR 10 instead of Replace at budget 100%posterior-based updates · ImageNet->ImageNet, budget 100%final accuracy (%): 79.2 -> 77Hurt
use CR 10 instead of Replace at budget 10% with entropy selectionposterior-based updates · ImageNet->ImageNet, budget 10%cumulative negative flips: 2468 -> 515Improved
use CR 10 instead of Replace at budget 10% with entropy selectionposterior-based updates · ImageNet->ImageNet, budget 10%final accuracy (%): 76.1 -> 75.2Hurt
select re-evaluation samples by posterior entropy instead of randomly (ablation)Replace strategy · ImageNet->ImageNet, budget 10%final accuracy (%): 71.3 -> 76.1Improved
select re-evaluation samples by posterior entropy instead of randomly (ablation)Replace strategy · ImageNet->ImageNet, budget 10%cumulative negative flips: 1958.4 -> 2468Hurt
use CR 10 instead of Replace on ObjectNet at budget 10%posterior-based updates · ImageNet->ObjectNet, budget 10%final accuracy (%): 23.7 -> 17.2Hurt
use CR 10 instead of Replace on ObjectNet at budget 100%posterior-based updates · ImageNet->ObjectNet, budget 100%cumulative negative flips: 16669 -> 1084Improved
estimate full Laplace-smoothed confusion matrices instead of diagonal only (ablation)posterior-based updates · ImageNet and ObjectNetnegative flipsImproved
estimate full confusion matrices instead of diagonal only (ablation)conservative CR strategies at small budget · ImageNet and ObjectNetaccuracy gainMixed

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