AI Paper Atlas

Research map / Continual learning / Class-incremental

On the Approximation Risk of Few-Shot Class-Incremental Learning

Xuan Wang, Zhong Ji, Xiyao Liu, Yanwei Pang, Jungong Han

ECCV 2024 · Continual learning › Class-incremental · approach: Few-shot class-incremental prototypes · 3 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
discard the base classifier and train per-task classifiers independently instead of fine-tuning the base classifier (ablation)FSCIL ViT-B/16 · CUB200final-session accuracy (%): 38.01 -> 70.97Improved
freeze the base classifier instead of fine-tuning it (ablation)FSCIL ViT-B/16 · CUB200final-session accuracy (%): 38.01 -> 52.41Improved
add tanh after each normalization layer (Stabilize) (ablation)ViT-B/16 with DS · CUB200DR (%): 18.47 -> 17.30Improved
add layer norm + tanh on shared representation (Constrain) (ablation)ViT-B/16 with DS · CUB200DR (%): 18.47 -> 16.69Improved
add label smoothing (ablation)ViT-B/16 with DS · CUB200DR (%): 18.47 -> 16.82Improved
combine Stabilize, Constrain and label smoothing (ablation)ViT-B/16 with DS · CUB200final-session accuracy (%): 70.97 -> 75.22Improved
use ImageNet-1K pretrained ViT instead of CLIP-pretrained ViTFSCIL ViT-B/16 · CUB200final-session accuracy (%): 75.22 -> 81.30Improved
increase base-class dataset size via data augmentation and MixupResNet18 and ViT-S from scratch · mini-ImageNetfinal accuracyImproved
apply the guidelines on mini-ImageNet and CIFAR100 with ViTsFSCIL ViT · mini-ImageNet, CIFAR100accuracyMixed

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