AI Paper Atlas

Research map / Continual learning / Class-incremental

RMM: Reinforced Memory Management for Class-Incremental Learning

Yaoyao Liu, Bernt Schiele, Qianru Sun

NeurIPS 2021 · Continual learning › Class-incremental · approach: Exemplar-efficient class-incremental learning

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
Plug RMM into POD-AANetsPOD-AANets · CIFAR-100, N=5average accuracy (%): 66.61 -> 68.86Improved
Plug RMM into POD-AANetsPOD-AANets · CIFAR-100, N=25average accuracy (%): 62.63 -> 66.21Improved
Plug RMM into POD-AANetsPOD-AANets · ImageNet-Subset, N=25average accuracy (%): 72.18 -> 76.54Improved
Plug RMM into POD-AANetsPOD-AANets · ImageNet-Full, N=25average accuracy (%): 62.03 -> 63.93Improved
Plug RMM into LUCIR-AANetsLUCIR-AANets · ImageNet-Subset, N=25average accuracy (%): 68.07 -> 72.30Improved
Plug RMM into LUCIR-AANetsLUCIR-AANets · CIFAR-100, N=5average accuracy (%): 66.88 -> 68.42Improved
Increase number of incremental phases N from 5 to 25 (RMM gain grows)POD-AANets w/ RMM · CIFAR-100 and ImageNet-Subsetgain over baseline: 2.3% -> 3.6% (CIFAR-100), 2.1% -> 4.4% (ImageNet-Subset)Improved
Add Level-1 RL policy for old/new memory split (ablation)POD-AANets · CIFAR-100, N=5average accuracy (%): 66.61 -> 67.92Improved
Add Level-2 class-specific policy (two-level instead of one-level) (ablation)RMM · CIFAR-100, N=5average accuracy (%): 67.92 -> 68.86Improved
Add Level-2 policy (two-level vs one-level) (ablation)RMM · ImageNet-Subset, N=25average accuracy (%): 75.47 -> 76.54Improved
Transfer policy from another dataset instead of learning on target (ablation)RMM · CIFAR-100, N=5average accuracy (%): 68.86 -> 68.62No clear effect
Train RMM on target task with all data (UpperBound RL) instead of pseudo tasks (ablation)RMM · CIFAR-100, N=5average accuracy (%): 68.86 -> 70.00No clear effect

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