AI Paper Atlas

Research map / Continual learning / Class-incremental

Prototype-Sample Relation Distillation: Towards Replay-Free Continual Learning

Nader Asadi, MohammadReza Davari, Sudhir Mudur, Rahaf Aljundi, Eugene Belilovsky

ICML 2023 · Continual learning › Class-incremental · approach: Class-balanced knowledge retention · 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
add prototype-sample relation distillation loss (beta>0) vs none (beta=0) (ablation)PRD · Split-CIFAR100 K=20 task-incrementalaverage observed accuracy: 39.4 -> 83.5 at beta=4Improved
add prototype-sample relation distillation loss (beta>0) vs none (beta=0) (ablation)PRD · Split-MiniImageNet K=20 task-incrementalaverage observed accuracy: 31.2 -> 70.9 at beta=8Improved
add prototype-sample relation distillation loss (beta>0) vs none (beta=0) (ablation)PRD · ImageNet32 K=200 task-incrementalaverage observed accuracy: 21.3 -> 67.5 at beta=16Improved
increase relation distillation coefficient beta (ablation)PRD · Split-CIFAR100 K=20 task-incrementalaverage observed accuracy: 80.0 to 83.5 across beta=1 to 4, 82.7 at 16No clear effect
add prototype learning loss (alpha>0) vs none (alpha=0) (ablation)PRD · Split-CIFAR100 K=20 task-incrementalaverage observed accuracy: 20.6 -> 83.5 at alpha=4Improved
add prototype learning loss (alpha>0) vs none (alpha=0) (ablation)PRD · Split-MiniImageNet K=20 task-incrementalaverage observed accuracy: 20.2 -> 69.8 at alpha=2Improved
vary prototype coefficient alpha in {1,2,4,8,16} (ablation)PRD · Split-CIFAR100 and Split-MiniImageNet K=20 task-incrementalaverage observed accuracy: 81.8-83.5 (CIFAR100), 67.4-69.8 (MiniImageNet) for alpha>=1No clear effect
use PRD (replay-free) instead of replay-based ER with 50 samples per classSplit-CIFAR100 K=20 task-incrementalaverage observed accuracy: PRD outperforms ER M=50 without stored dataImproved
add a small replay buffer (M=5/20/50) to PRDPRD · Split-CIFAR100 K=20 class-incrementalaverage observed accuracy: 27.8 (M=0) -> 32.0 (M=5) -> 39.5 (M=20) -> 45.1 (M=50)Improved
replay-free PRD vs replay-based methods in class-incrementalPRD · Split-CIFAR100 K=20 class-incrementalaverage observed accuracy: 27.8 (M=0) vs ER M=5 15.2Improved
replay-free PRD vs SPB and distillation baselines with pre-trained initializationPRD · Split-CIFAR100 and ImageNet-Subset pre-trained init class-incrementalaverage cumulative incremental accuracy: PRD 64.3 vs SPB 60.9 on Split-CIFAR100 K=6; 71.8 vs 68.7 on ImageNet-Subset K=6Improved
replay-free PRD vs EWC and LwF in domain-incremental CLAD-CPRD · CLAD-C domain-incrementalAMCA: 65.1 vs LwF 63.7 and EWC 62.5Improved

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