AI Paper Atlas

Research map / Continual learning / Class-incremental

Continual Learning with Evolving Class Ontologies

Zhiqiu Lin, Deepak Pathak, Yu-Xiong Wang, Deva Ramanan, Shu Kong

NeurIPS 2022 · Continual learning › Class-incremental · approach: Distillation-based continual segmentation · 2 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
finetune previous-TP model instead of training from scratchLabelNew · iNat-LECOmAcc: 65.69 -> 73.64Improved
freeze the previous-TP backbone instead of training from scratchLabelNew · iNat-LECOmAcc: 65.69 -> 50.82Hurt
relabel old data instead of labeling new dataFinetunePrev · iNat-LECOmAcc: 73.64 -> 71.26No clear effect
add ST-Soft semi-supervised loss on old dataFinetunePrev LabelNew · iNat-LECOmAcc: 73.64 -> 79.64Improved
add joint training loss using old coarse labelsFinetunePrev LabelNew · iNat-LECOmAcc: 73.64 -> 82.98Improved
combine Joint and ST-Soft lossesFinetunePrev LabelNew · iNat-LECOmAcc: 73.64 -> 83.61Improved
add Joint loss on top of ST-SoftST-Soft · CIFAR-LECOmAcc: 69.86 -> 69.77No clear effect
add LPL loss using taxonomic hierarchyFinetunePrev LabelNew · iNat-LECOmAcc: 73.64 -> 83.62Improved
add SSL with conditioning pseudo-label refinement to LPL (ST-Hard)LPL · iNat-LECOmAcc: 83.62 -> 84.34Improved
add SSL with filtering refinement to LPL (ST-Hard)LPL · iNat-LECOmAcc: 83.62 -> 84.23Improved
add SSL/Cond to LPL (ST-Hard)LPL · CIFAR-LECOmAcc: 69.31 -> 70.42Improved
use Conditioning instead of Filtering with FixMatch (ablation)LPL + FixMatch · iNat-4TP-LECO TP3mAcc: 85.3 -> 78.8Hurt

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