AI Paper Atlas

Research map / Continual learning / Class-incremental

Remember the Past: Distilling Datasets into Addressable Memories for Neural Networks

Zhiwei Deng, Olga Russakovsky

NeurIPS 2022 · Continual learning › Class-incremental · approach: Class-balanced knowledge retention · 21 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 shared bases with learned addressing matrices (ablation)Ours (memory addressing + BPTT) · CIFAR10 1 img/classaccuracy (%): 55.2 -> 64.2Improved
add downsampled bases (ablation)Ours (memory addressing + BPTT) · CIFAR10 1 img/classaccuracy (%): 49.1 -> 55.2Improved
add data augmentation (ablation)Ours (memory addressing + BPTT) · CIFAR10 1 img/classaccuracy (%): 64.2 -> 66.4Improved
add momentum in inner loop (ablation)Ours (memory addressing + BPTT) · CIFAR10 1 img/class BPTTaccuracy (%): 40.5 -> 49.1Improved
pass momentum gradients through backward pass (full vs forward-only) (ablation)Ours (memory addressing + BPTT) · CIFAR10 1 img/class BPTTaccuracy (%): 45.6 -> 49.1Improved
use longer unrolled inner loop (vs 1 step) (ablation)Ours (memory addressing + BPTT) · CIFAR10 BPTTaccuracy (%): +18.2Improved
use BPTT instead of single-step gradient matching (ablation)Dataset distillation · CIFAR10 1 img/classaccuracy (%): 28.8 -> 49.1Improved
use Adam instead of SGD with momentum for inner loop (ablation)Ours (memory addressing + BPTT) · distillationaccuracy (%)No clear effect
distill with memory addressing instead of per-task buffer for continual learningcompress-then-recall · MANY Permutationsretained accuracy: 50.81 -> 74.07Improved
address memories with image features instead of labels (vs nearest neighbor)Ours (memory addressing + BPTT) · CIFAR100 5-shotaccuracy (%): 61.72 -> 71.20Improved
image addressing vs classify-then-recallOurs (memory addressing + BPTT) · CIFAR100 5-shotaccuracy (%): 58.46 -> 71.20Improved

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