AI Paper Atlas

Research map / Continual learning / Online and replay-based

Smart Replay: Adaptive Scheduling of Memory Rehearsal for Computational Resource-Aware Incremental Learning

Jianting Chen, Dianzhi Yu, Irwin King

CVPR 2026 · Continual learning › Online and replay-based · approach: Replay buffer construction and update

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
schedule replay ratio adaptively with Smart Replay instead of fixed ratioER · CIFAR-100 limited memory, ϑ=0average accuracy (%): 63.84 -> 64.58 vs best linear heuristicImproved
schedule replay ratio adaptively with Smart ReplayiCaRL · Tiny-ImageNet limited memory, ϑ=0.1average accuracy (%): 44.80 -> 49.08Improved
schedule replay ratio adaptively with Smart ReplayiCaRL · Rotated-MNIST limited memory, ϑ=0.2average accuracy (%): 90.93 -> 90.69 vs fixed λ=0.4No clear effect
schedule replay ratio adaptively with Smart Replay on unlimited memoryER · CIFAR-100 unlimited memory, ϑ=0average accuracy (%): 74.16 -> 74.70 vs UnionImproved
decreasing replay ratio schedule (0.8 to 0.2) instead of increasingER · CIFAR-100 limited memory, ϑ=0average accuracy (%): 63.84 -> 57.91Hurt
add smoothness regularization on replay ratio changes (ablation)Smart Replay · CIFAR-100 limited memoryaverage accuracy (%): 36.89 -> 64.58Improved
add short-term gradient norm ratio term (ablation)Smart Replay · CIFAR-100 limited memoryaverage accuracy (%): 40.20 -> 64.58Improved
add long-term weight ratio term (ablation)Smart Replay · CIFAR-100 limited memoryaverage accuracy (%): 61.43 -> 64.58Improved
add gradient-direction cosine term (ablation)Smart Replay · CIFAR-100 limited memoryaverage accuracy (%): 64.03 -> 64.58Improved
increase decay-rate difference Δρ from 1.0 to 3.0 (ablation)Smart Replay · CIFAR-100 limited memoryaverage accuracyImproved
vary update step size γ, interval z and initial ratio τ (ablation)Smart Replay · CIFAR-100average accuracyNo clear effect
use larger fixed replay ratio with limited bufferER · Tiny-ImageNet limited memory, ϑ=0.2average accuracy (%): 43.96 -> 41.20 (λ=0.2 -> 0.4)Hurt

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