AI Paper Atlas

Research map / Continual learning / Class-incremental

Efficient Feature Transformations for Discriminative and Generative Continual Learning

Vinay Kumar Verma, Kevin J Liang, Nikhil Mehta, Piyush Rai, Lawrence Carin

CVPR 2021 · Continual learning › Class-incremental · approach: Class-balanced knowledge retention · 62 citations (OpenAlex)

Check 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
replace prior continual-learning methods with EFT feature transformsResNet-18 · CIFAR-100/10 CILfinal average accuracy: 36.2 -> 45.5Improved
replace PackNet with EFT feature transformsVGG-16 · Tiny-ImageNet-200/10 task incrementalaverage top-1 accuracy: 56.4 -> 66.8Improved
replace PackNet with EFT feature transformsAlexNet · ImageNet-1K/10 task incrementalaverage top-1 accuracy: 59.1 -> 59.4Improved
add feature distance regularization (ablation)EFT-a8b16 · CIFAR-100/10CIL accuracy: 44.4 -> 45.5Improved
add feature distance regularization (ablation)EFT-a8b16 · CIFAR-100/10task prediction accuracy: 44.9 -> 46.1Improved
increase transform group sizes from a4b0 to a8b16 (ablation)EFT · CIFAR-100/10accuracy: 44.4 -> 45.5Improved
use spatial transforms alongside pointwise transforms rather than pointwise transforms alone (ablation)EFT · CIFAR-100/10accuracy: 43.2 -> 45.5Improved
initialize each new task's transform from the previous task's transform instead of randomly (ablation)EFT-a8b16 · CIFAR-100/10accuracyImproved
use parallel rather than sequential feature adaptation (ablation)EFT · continual learning classificationaccuracyHurt
use EFT instead of MeRGAN-RAStackGAN-v2 · continual generation: cats, birds, churches, towersfinal average FID: 42.8 -> 33.1Improved
replace APD with EFTVGG-16 · heterogeneous sequence SVHN → CIFAR-10 → CIFAR-100average accuracy: 83.0 -> 83.5Improved
replace APD with EFTVGG-16 · heterogeneous sequence CIFAR-100 → CIFAR-10 → SVHNaverage accuracy: 85.0 -> 85.8Improved

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