AI Paper Atlas

Research map / Continual learning / Class-incremental

Exemplar-free Continual Representation Learning via Learnable Drift Compensation

Alex Gomez-Villa, Dipam Goswami, Kai Wang, Andy Bagdanov, Bartlomiej Twardowski, Joost van de Weijer

ECCV 2024 · Continual learning › Class-incremental · approach: Class-balanced knowledge retention · 15 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 learnable drift compensation (LDC) to LwF+NCM prototypesLwF with NCM classifier · CIFAR-100 (10 tasks, cold start)last-task accuracy: 40.5 -> 45.4Improved
replace SDC with LDC for prototype correctionLwF · ImageNet100 (10 tasks, cold start)last-task accuracy: 42.6 -> 51.4Improved
add SDC prototype correction to LwF+NCM in cold startLwF with NCM classifier · CIFAR-100 (10 tasks, cold start)last-task accuracy: 40.5 -> 40.6No clear effect
add LDC to fine-tuning with pretrained ViT-B/16FT+NCM · Stanford Cars (10 tasks)accuracy after task 10: 53.4 -> 62.9Improved
add LDC to self-supervised CL methods for semi-supervised exemplar-free learningCaSSLe · CIFAR-100 semi-supervised (25% labels)last-task accuracy: 23.6 -> 35.0Improved
add LDC to PFR in semi-supervised setting with very few labelsPFR · CIFAR-100 semi-supervised (0.8% labels)last-task accuracy: 20.5 (naive) -> 18.6Hurt
use a single linear projector instead of ReLU after the linear layer (ablation)LwF+LDC · CIFAR-100 (10 tasks)last-task accuracy: 41.2 -> 45.4Improved
use a single linear projector instead of a two-layer MLP (ablation)CaSSLe+LDC · CIFAR-100 semi-supervised (25% labels)last-task accuracy: 34.9 -> 35.0No clear effect
add bias to the linear projector (ablation)LwF+LDC · CIFAR-100 (10 tasks)last-task accuracy: 45.4 -> 45.2No clear effect
project stored per-class features instead of class means (ablation)CaSSLe+LDC · CIFAR-100 semi-supervised (25% labels)last-task accuracy: 35.0 -> 35.05 (N=500)No clear effect
give NNCSL exemplars versus LDC without exemplarsexemplar-free LDC vs NNCSL · ImageNet100 semi-supervised (25% labels)last-task accuracy: NNCSL (5120) 47.1 vs CaSSLe+LDC 35.8Hurt
replace exemplar-based NME with LDC for prototype correction (ablation)LwF+NCM · CIFAR-100, Tiny-ImageNet, ImageNet100last-task accuracyNo 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