Research map / Continual learning / Class-incremental
AREA: Attribute Extraction and Aggregation for CLIP-Based Class-Incremental Learning
Zhen-Hao Xie, Yu-Cheng Shi, Da-Wei Zhou
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add PGA-anchored attributes with multi-modal aggregation to zero-shot CLIP (ablation) | AREA · CIFAR100 B0 Inc10 | accuracy | Improved |
| add VIB loss (intervention and compression terms) (ablation) | AREA · CIFAR100 B0 Inc10 | accuracy | Improved |
| add optimal transport routing at inference (ablation) | AREA · CIFAR100 B0 Inc10 | accuracy | Improved |
| use OT routing instead of cosine routing (ablation) | AREA · CIFAR100 B0 Inc10, final stage | final accuracy: +3.39% accuracy, +2.9 ms/sample | Improved |
| use PGA anchors instead of SVD anchors (ablation) | AREA attribute extraction · ObjectNet attribute similarity | similarity to fine-grained descriptions | Improved |
| caption only 5% to 20% of training data with MLLM (ablation) | AREA · CIFAR100 | accuracy: 87.9% at 5%; drop under 1% from 100% to 20% | No clear effect |
| swap captioning MLLM from GPT-5 to LLaVA-v1.6-34b (ablation) | AREA · Aircraft, CIFAR, CUB | average accuracy: 71.03 -> 70.89 (Aircraft); 89.24 -> 88.98 (CIFAR) | No clear effect |
| vary lambda_int and lambda_comp in 0.5 to 1.5 (ablation) | AREA · CIFAR100 | final average accuracy: range 87.86 to 89.24 | No clear effect |
| AREA versus MG-CLIP on CIFAR average accuracy | CLIP-based CIL · CIFAR B0 Inc10 | average accuracy: 89.74 (MG-CLIP) vs 89.24 | Hurt |
| AREA versus RAPF on hard datasets | CLIP-based CIL · Aircraft B0 Inc10 | last accuracy: 23.61 -> 61.78 | Improved |
| add attribute anchors to prevent extraction drift (ablation) | AREA · SUN example | prediction rank of correct class | Improved |
Closest papers by meaning
- eTag: Class-Incremental Learning via Embedding Distillation and Task-Oriented Generation (AAAI 2024 · similarity 0.87)
- Navigating Semantic Drift in Task-Agnostic Class-Incremental Learning (ICML 2025 · similarity 0.87)
- Task-Agnostic Guided Feature Expansion for Class-Incremental Learning (CVPR 2025 · similarity 0.86)
- Hierarchical Semantic Tree Anchoring for CLIP-Based Class-Incremental Learning (NeurIPS 2026 · similarity 0.86)
- Task-recency bias strikes back: Adapting covariances in Exemplar-Free Class Incremental Learning (NeurIPS 2024 · similarity 0.86)
- Reducing Class-wise Confusion for Incremental Learning with Disentangled Manifolds (CVPR 2025 · similarity 0.86)
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.
