AI Paper Atlas

Research map / Continual learning / Class-incremental

Online Class Incremental Learning on Stochastic Blurry Task Boundary via Mask and Visual Prompt Tuning

Jun-Yeong Moon, Keon-Hee Park, Jung Uk Kim, Gyeong-Moon Park

ICCV 2023 · Continual learning › Class-incremental · approach: Class-balanced knowledge retention · 30 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
add instance-wise logit masking to DualPrompt baseline (ablation)DualPrompt · Si-Blurry CIFAR-100 (buffer 0)A_Last: 56.82 -> 58.52Improved
add contrastive visual prompt tuning loss to DualPrompt baseline (ablation)DualPrompt · Si-Blurry CIFAR-100 (buffer 0)A_AUC: 67.07 -> 66.37 (A_AUC); 56.82 -> 58.63 (A_Last)Mixed
combine logit masking and contrastive loss (ablation)MVP · Si-Blurry CIFAR-100 (buffer 0)A_Last: 58.52 -> 60.20Improved
add GSF loss alone (ablation)DualPrompt · Si-Blurry CIFAR-100 (buffer 0)A_Last: 56.82 -> 56.11Hurt
add adaptive feature scaling alone (ablation)DualPrompt · Si-Blurry CIFAR-100 (buffer 2000)A_Last: 80.72 -> 84.31Improved
add all four components (Mask, Cont, GSF, AFS) (ablation)DualPrompt · Si-Blurry CIFAR-100 (buffer 0)A_Last: 56.82 -> 62.59Improved
add replay buffer of 2,000 samples (MVP-R vs MVP)MVP · Si-Blurry CIFAR-100A_Last: 62.59 -> 84.42Improved
gamma in GSF loss set to 2.0 versus 0.5 (ablation)MVP · Si-Blurry CIFAR-100A_Last: 60.39 -> 62.59Improved
MVP versus DualPrompt when only disjoint classes presentMVP · Si-Blurry CIFAR-100 (disjoint ratio 0)A_AUC: 68.85 -> 67.86Hurt
raise blurry sample ratio to 50MVP · Si-Blurry CIFAR-100A_Last: 49.93 -> 59.47Improved
Si-Blurry versus i-Blurry for MVP-RMVP-R (2,000) · CIFAR-100A_Last: 84.44 vs 84.42No clear effect

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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