AI Paper Atlas

Research map / Continual learning / Class-incremental

Few-Shot Continual Active Learning by a Robot

Ali Ayub, Carter Fendley

NeurIPS 2022 · Continual learning › Class-incremental · approach: Class-balanced knowledge retention · 12 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 GBCL rather than random sampling (ablation)CORe-50 FoCAL · CORe-50accuracyImproved
Combine predictive entropy and viewpoint consistency rather than use either acquisition function alone (ablation)GBCL active learning · Pepper datasetaccuracy: only 0.07%Improved
Set the entropy weight to 0.7 rather than use either acquisition function alone (ablation)GBCL active learning · Pepper datasetaccuracy: 0.7; only 0.07%Improved
Use a probability threshold near 0 rather than the selected threshold (ablation)GBCL GMM clustering · Pepper datasetaccuracy: ∼18%Hurt
Increase the GMM probability threshold to 0.15 (ablation)GBCL GMM clustering · Pepper datasetaverage incremental accuracy: P=0.15; 41 total clustersImproved
Increase the GMM probability threshold to 0.2 (ablation)GBCL GMM clustering · Pepper datasetaccuracy: P=0.2; 83 total clustersImproved
Increase the GMM probability threshold to 0.7 (ablation)GBCL GMM clustering · Pepper datasetaverage incremental accuracy: P=0.7; 379 clusters; 4.5 timesImproved
Use a probability threshold near 1, recruiting many clusters with few features each (ablation)GBCL GMM clustering · Pepper datasetstorage useHurt
Use GBCL rather than FLBCORe-50 FoCAL · CORe-50, 400 incrementsaccuracy: ∼89% vs ∼88%; 1% lowerHurt
Use GBCL rather than CBCLCORe-50 FoCAL · CORe-50accuracyImproved
Use GMM-based active learning rather than random sampling (ablation)GBCL · CORe-50, early incrementsaccuracyImproved
Use a fixed feature extractor rather than adapt the representation during incremental learningGBCL vs TOPIC · CIFAR-100 FSCILaccuracy: ∼8%Mixed

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