AI Paper Atlas

Research map / Continual learning / Online and replay-based

Gradient-based Editing of Memory Examples for Online Task-free Continual Learning

Xisen Jin, Arka Sadhu, Junyi Du, Xiang Ren

NeurIPS 2021 · Continual learning › Online and replay-based · approach: Replay buffer construction and update · 39 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 GMED editing to memory replayMIR · Split MNISTfinal accuracy (%): 85.72 ± 1.2 -> 86.52$^{**}$ ± 1.4Improved
Add GMED editing to memory replayMIR · Permuted MNISTfinal accuracy (%): 79.13 ± 0.7 -> 79.25 ± 0.8No clear effect
Add GMED editing to augmented memory replayER_aug · Split CIFAR-10final accuracy (%): 46.29 ± 2.7 -> 47.47$^*$ ± 3.2Improved
Replace the GMED objective with random fixed-stride perturbations (ablation)ER replay · Split mini-ImageNetfinal accuracy (%): 25.92 ± 1.2 -> 26.50 ± 1.3Improved
Use GMED's interference-focused edit rather than random perturbation (ablation)ER replay · Split CIFAR-10final accuracy (%): 32.26 ± 1.9 -> 34.84$^*$ ± 2.2Improved
Use GMED's interference-focused edit rather than adversarial loss maximization (ablation)ER replay · Split CIFAR-10final accuracy (%): 31.69 ± 0.8 -> 34.84$^*$ ± 2.2Improved
Increase editing from one to three gradient steps (ablation)ER_aug + GMED · Split CIFAR-10final accuracy (%): 34.84 ± 2.2 -> 34.93 ± 1.4No clear effect
Increase editing from one to five gradient steps (ablation)ER_aug + GMED · Split mini-ImageNetfinal accuracy (%): 27.27 ± 1.8 -> 26.36 ± 2.0Hurt
Use a nonzero regularization strength to constrain editing (ablation)ER + GMED · Across tested datasetsfinal accuracyImproved
Set editing decay rate γ to 1.0 instead of less than 1 (ablation)ER + GMED and MIR + GMED · Across six datasetsfinal accuracy (%)Mixed
Replay additional memory batches instead of editing examples (ablation)MIR · Rotated MNISTfinal accuracy (%): 77.50 ± 1.6 -> 78.07 ± 1.1Mixed
Edit memory examples using GMEDER + GMED · Fuzzy-boundary Split CIFAR-10final accuracy (%): 37.15 ± 1.6 -> 40.57$^*$ ± 1.7Improved

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